Introduction

Dogs and man have a complex relationship going back many thousands of years [1]. Dogs, as one of the most widely distributed carnivores, also interact with wildlife in various environments across rural-urban gradients. They often adopt the role of predators [2,3], prey [45–6], competitors [7] and reservoirs or transmitters of disease [8,9]. Additionally, they may also take up the role of “conservation dogs” (sensu Reed et al. [10]), and are employed in wildlife research and management [11]. Dogs can broadly be categorized as owned, urban or rural free-ranging, feral or village (as per [12]). Free-ranging dogs often come into contact with wildlife and interact negatively [13,14], sometimes even outcompeting native species [15], as well as posing a high risk of disease spillover to native carnivores, especially to wild canids and felids [5]. Studies from around the globe highlight the influence of free-ranging dogs on native fauna [3,5,1617–18].

Understanding dog abundance is a fundamental requirement in most dog control and management strategies, whether it involves euthanasia, sterilization programmes or other approaches. A number of methods have been suggested to enumerate dog populations, including the use of block counts [19], mark-recapture using paint solution sprays [20], household surveys [21], collar marking with transect line recaptures [22], and distance sampling [23]. Recently, the mark-resight framework has been applied to estimate dog abundance using ear-notched marks made during Animal Birth Control (ABC) programmes [17]. These methods have been used to understand dog populations covering a range of spatial scales from rural and peri-urban areas, to the level of city blocks and even entire cities. They can be employed in varied circumstances of time, monetary and logistical availability; however many of them require artificial marking. We here describe a potential method, based on the results of a pilot study, which exploits the use of natural marks present on individual dogs to estimate their abundance in a mark-resight framework. This method may be a useful addition to the suite of modeling methods available for counting dogs and can be used at variable scales, especially in the absence of artificial marks.

In many free-ranging dog populations, some dogs may have natural marks by which individuals can be easily identified visually, while others may be mono-coloured or not distinctly marked, which makes it difficult to identify them visually as unique individuals. In such situations, natural marks can be employed to estimate the abundance of a closed free-ranging dog population in a mark-resight framework, given that some assumptions are met. The primary assumption is that the marked population should be representative of the unmarked population in terms of sightability [24]. This is a reasonable assumption with domestic animals that are generally highly visible and not shy or elusive. In normal circumstances, dog sightability also should not be influenced by the presence or absence of natural marks. This is important since naturally marked individuals can then be assumed to belong to a random subset of the population. Other important assumptions of the mark-resight approach include geographic and demographic closure during primary sampling intervals and no loss or change of marks [25]. Geographic closure may be relaxed in some circumstances; however, the estimate of abundance is then termed a ‘super population’ [25].

Population estimates are generally more sensitive to heterogeneity, which can result in biased estimates of abundance and over-precision in mark-resight studies [25,26]. Therefore, we used the logit-normal estimator which accounts for individual sighting heterogeneity, since we suspected differential sighting probabilities among dogs. Also, sighting probabilities for marked and unmarked individuals should be independent, and identically distributed [25]. This may be an issue with free-ranging dogs which inherently form loose aggregations; therefore sighting probabilities may not always be independent, leading to underestimation of parameter variance. However, similar mark-resight estimators used with group-living animals indicate that biases associated with parameter estimates are low [27282930–31]; therefore, we believe that our estimate of population size is still useful. Lastly, it is important to perform the marking or identification of marked individuals immediately before actual sampling, in order to know the exact number of marked individuals present during sampling.

This pilot study of dog ecology was prompted by an ongoing conflict study in the area, where leopards (Panthera pardus) leave a neighbouring national park to enter suburban environments in pursuit of dogs, which are a major prey item [4]. This brings leopards into potential conflict situations with humans, which can occasionally have fatal results for people.

Methods

Dog surveys

We conducted the entire study from 25^th February to 1^st March, 2012. We initially surveyed all of AMC on a motorbike and identified points where dogs were observed, either in clusters or singly. We tracked the majority of the roads within the AMC using a hand-held GPS (Garmin Inc., USA, GPS 72H) and marked all locations with dog presence. These points were generally close to human habitations, and included garbage and cattle carcass dumps, cattle units, slums, villages, and the premises of institutions within the AMC. A total of 65 such points were identified and overlaid on a map of the AMC to design a survey route. Although we recorded areas where dogs were observed, we ensured that our survey route uniformly covered most areas within the colony where dogs could be found. However, we did not cover a small portion (c. 1.3 km²) in the western edge of the AMC due to its inaccessibility and time constraints (Fig. 1). We ensured that the area covered was large and that the perimeter to area ratio was small, so as to avoid violation of the assumption of geographic closure. Since urban/ village dogs are known to have small home ranges and high territoriality ([16] and references therein) we believe that our study satisfied the assumption of geographic closure for the short period of study.

Fig. 1.

The map of the study area indicates the survey route and dog count points used to estimate dog abundance in a mark-resight framework in Aarey Milk Colony, India.

We first followed this survey route to identify and photograph free-ranging adult dogs with easily distinguishable natural marks. The majority of dogs were strays, while some dogs may be categorized as semi-owned, since they were associated with specific cattle units or households; all the dogs were unrestricted by humans in their movements. Dogs were identified as marked by observing natural marks on their flanks, head, and tail. We ensured that during our initial survey we identified as many naturally marked dogs as we could detect. Such naturally marked dogs were identified as marked individuals in the population. This was done two days prior to actual sampling to adhere to the assumption of closure, since mortality rates of free-ranging dogs are assumed to be moderate to high [17,33]. Mono-coloured adult dogs or dogs without conspicuous natural marks were regarded as unmarked individuals in the population. We also observed that free-ranging dogs were most visible during morning and evening hours in our study area. Thus, surveys (initial and actual) were planned keeping in mind activity and visibility levels of dogs, in order to increase the sightability of individuals.

We conducted the primary sampling surveys during daylight hours over three days, where each day comprised one secondary sampling interval. The primary sampling interval (28^th February to 1^st March 2012) was kept short to adhere to the assumptions of geographic and demographic closure. On each secondary sampling interval, the entire survey route was covered using a motorbike and high quality photographs of dogs with natural marks were taken using a digital SLR camera (Canon India Pvt. Ltd., India, Canon 1000D, Canon 18–55 mm lens) (Fig. 2). The photographs of dogs included as many distinct natural marks as possible on each individual to minimize the possibility of misidentification. At each point, as well as along the survey route, all marked dogs seen were photographed and the numbers of unmarked dogs were recorded. We thoroughly surveyed the area around each point to detect as many dogs as possible, ensuring that we did not traverse the surveyed area twice to avoid double counts. The survey route was covered in a unidirectional way such that sampling was without replacement, i.e., each individual was sighted only once during each secondary sampling interval [34]. If some roads had to be repeatedly traversed on the same sampling interval to reach other points, we did not count or photograph dogs during this repeat traverse to avoid double counts of unmarked dogs.

The logit-normal mark resight estimator may be employed when the number of marked individuals in the population is known exactly and sampling is without replacement [34]. The model provides an estimate of abundance (N_j) for primary interval j, mean resighting probability (p_ij) for each secondary sampling interval i of primary interval j based on the kind of model used, and individual heterogeneity parameter (σ_j), when individual identification is possible [25]. When heterogeneity is estimated, the model also provides an estimate of overall mean resighting probability (µ_ij), which otherwise is equal to p_ij. For our study, we photographed all individuals with distinct natural marks, as it was essential to identify whether the individual was from the known marked population, or an entirely new individual with distinct natural marks, in order to avoid biasing the overall mean resighting probability (µ_ij). Such naturally marked individuals, which were not identified during our initial surveys, were included in the unmarked population. Although our goal was to estimate the abundance (N_j) of dogs, this method can be used to estimate other demographic parameters of dog populations as well. This is possible when data are collected over multiple primary intervals, and new marks may be introduced at any time between sampling intervals (refer to [25,35] for details).

Data input and Analysis

We performed the analysis in Program MARK ([36], Version 6.2) using the mixed logit-normal mark-resight estimator, where the number of marked individuals in the population should be known exactly [34]. This is similar to Bowden's estimator in the NOREMARK program [24]. As marks were individually identifiable, we used individual heterogeneity models with the data, since we expected individual dogs to have differential sighting probabilities. We used the time-constant models, with or without individual heterogeneity parameters. Since some convergence issues occurred for values of s (individual heterogeneity parameter) for the time-constant model with heterogeneity, we manually supplied initial values covering the probable range of parameters in the three models. We also used other options to overcome convergence issues for values of s, such as the ‘Alt.Opt.method’ and ‘Do not standardize design matrix’ in another two models. In all we used seven time-constant models, the first six of which represented the time-constant model with individual heterogeneity, with various options as described above. The last model represented the time-constant model without the individual heterogeneity parameter (where s was fixed to zero). Model comparison was performed in an information-theoretic framework using Akaike's information criteria corrected for small sample size (AICc).

Fig. 2.

The image indicates photographs obtained for a distinct naturally marked dog over two secondary sampling intervals in Aarey Milk Colony, India.

Results

We identified a total of 101 naturally marked individuals in the population before actual sampling, and these were resighted over three secondary sampling intervals during the primary interval. Fifty-seven dogs were resighted once, 29 dogs were resighted twice and only 15 dogs were resighted all three times across the three sampling intervals. A total of 270, 294 and 356 unmarked dogs were counted over the three secondary sampling intervals.

We used a total of seven models, of which the first six represented the time-constant individual heterogeneity model and one model represented the time-constant model where (individual heterogeneity parameter) s was fixed to zero. All seven models were ranked evenly, since s could not be estimated correctly in any of the first six models and experienced convergence problems. All real parameter estimates of N_j (Abundance) and p_ij (mean resighting probability) were exactly identical and there was no difference in the AICc scores for any model.

A total abundance (N_j) of 680.64 ± 34.06 (95% CI = 617.22 − 751.35) dogs were estimated to be present in the AMC. Mean resighting probability (p_ij) and overall mean resighting probability (µ_ij) was estimated to be 0.53 ± 0.03 (95% CI = 0.47 − 0.58).

Discussion

We demonstrate a simple approach for estimating dog numbers using natural marks with the flexible logit-normal mark-resight model [34]. The method has several advantages, especially in reducing cost for marking individuals, which has usually been recognized as the most expensive component of mark-resight studies [37]. Using natural marks avoids the need of artificially marking dogs or handling them for the purpose of marking. This reduces the risk of stray dog bites or disease exposure to researchers, as well as possible biases due to dog shyness as a result of handling or marking. When used over longer sampling intervals, this method may be useful to estimate abundance at smaller spatial scales, such as the level of blocks, wards, or villages, and may also be useful at the scale of larger cities, where randomly selected spatial sub-units (such as blocks) can be sampled. However, given the need to identify a fair proportion (at least 15%) of the population as ‘marked’ before sampling, the method may have potential limitations in covering relatively large populations occurring at larger spatial scales. Although we could not assess potential biases in this pilot study, we refer the reader to McClintock et al. [34] to understand the range of scenarios under which the logit-normal estimator may be efficient. We recognize that our model parameter variances may be underestimated since we could not account for possible overdispersion, due to slight non-independence among marked individuals. However, we would like to point out that similar mark-resight estimators used with group-living animals indicate that biases associated with parameter estimates are low [27282930–31], therefore our method and estimate of population size are useful.

Implications for conservation

Since free-ranging dogs are a major conservation issue in the tropics, a brief overview of some other methods available to enumerate dog populations would be useful. Hiby et al. [17] demonstrated a convenient mark-resight approach to estimate dog abundance for entire cities using ear-notched marks made during animal birth control (ABC) initiatives. Totton et al. [20] used paint sprays to mark individuals for estimation of dog population in a mark-recapture framework at the scale of areas used for ABC initiatives. Childs et al. [23] demonstrated the use of distance sampling for estimation of dog densities, when assumptions of mark-recapture methods were not met. Distance sampling can enable larger area coverage due to logistic convenience, but may not be useful when parameters such as individual heterogeneity or survival are of interest. Kitala et al. [21] and Matter et al. [22] used questionnaire and household surveys to estimate numbers of owned dogs. Household surveys are applicable when other approaches may be unsuitable to enumerate such ‘confined’ dog populations.

Our method may be appropriate when point estimates of free-ranging dog populations are required, but no readily available marks exist and limitations preclude other forms of artificial marking. The establishment of photographic databases of individual animals and computer-aided photo-recognition software could facilitate upscaling to larger survey areas. At similar scales, our method should be applicable in counting other species of urban or feral animals such as cats (Felis catus) and pigs (Sus scrofa), where some proportion of individuals in a population have distinct natural marks. Because such feral species are often subject to control programmes, understanding the abundance of feral populations of animals before an intervention and successively over the years will help evaluate the success of such initiatives. Our method accounts for imperfect detectability and individual sighting heterogeneity; not accounting for it may result in biased and over-precise estimates of abundance [34]. Assessing bias and precision of this method at relatively larger population sizes and proportions of naturally-marked individuals in the population requires further exploration.

The dog to man ratio estimated in India is about 1:36, and stray dogs are responsible for most bites causing human rabies [38]. Human rabies causes more deaths in India than in any other country [39]. Dog management is poor and the majority of dogs, even those with a formal owner, are partially or totally unrestricted in their movements [40]. In our study area (c.12 km²) in suburban Mumbai, we estimated a total of 681 (95% CI = 617 – 751) free-ranging dogs, which would yield densities of approximately 57 (51 – 63) dogs km⁻². All dogs were mixed-bred mongrels, and the majority of them were strays. No dogs had any collar to indicate ownership, and even those dogs that seemed semi-owned were unrestricted by humans in their movements. The results of our pilot study demonstrate a reasonably large population of free-ranging dogs in the AMC of suburban Mumbai, acting as potential prey for leopards as well as posing a disease risk to humans. Currently in our study area, systematic Animal Birth Control initiatives, as well as vaccination drives, are recommended, in order to control the dog population and minimize disease risk to humans. However, there is no established site-specific monitoring of the effect of control programs on dog populations over time. Our study provides a practical and time-effective method to assess changes in dog populations in such areas where control programmes are in progress.