Home range size and habitat selection of owned outdoor domestic cats (Felis catus) in urban southwestern Ontario

Study Site

Movement data on owned cats with unsupervised outdoor access were collected from Aug.–Nov. 2019 and May–Oct. 2021 in the Kitchener-Waterloo-Cambridge-Guelph region of southwestern Ontario (Fig. 1). The cities within this region contain over 1000 ha of parklands and have small urban cores, with a combined human population of 816,873 (Statistics Canada, 2021a; Statistics Canada, 2021b). Land cover in the region is made up of agricultural land, commercial land, natural open spaces, and mixed residential houses with most of the population residing in single-detached houses (Statistics Canada, 2021a; Statistics Canada, 2021b). Proportions of this text were previously published as part of a preprint (Pyott, 2023).

Equipment

As part of a larger study, we collected location and video data from outdoor cats by attaching a GPS and camera to a cat break-way collar. For the purpose of this study, we solely focused on the GPS data. The collars (GPS, 35 g; Catcam, 70 g; collar, 9 g), which weighed less than ∼5% of the smallest cat’s body weight (range: 2.3–8.6 kg), had an animal-borne camera on the front that rested below the cat’s chin, and a GPS unit on the back/top of the collar, that rested on the neck (Fig. 2; Tractive models TRATR3G and TRNJA4; Pasching, Austria). Owners were provided with the equipment for 5 wks and asked to collect at least 20 d of their cat’s outdoor activity. We asked owners to collect 20 d of data within a 5 wk period to increase the likelihood of compliance with the protocol and meet human ethics requirements by ensuring study participants had flexibility and did not feel forced to participate everyday. Despite the GPS having a standard battery life of 2–4 d, we instructed owners to charge the equipment for ∼2 hrs/d. The GPS units provided a fix of the cat’s location every 2–60 min, depending on the activity level of the cat and, in ideal conditions, had a positional accuracy of 8 m based on manufacturer details (Tractive, 2021) but with field testing accuracy was 4 ± 10 m in partial shade and 4.8 ± 1 m in areas fully exposed to the sky. All methods were approved by the University of Guelph’s Research Ethics Board (approval #4189), and the University of Guelph’s Animal Care Committee (AUP #4183).

Participants

We recruited cat owners by advertising through email listservs, social media, and news reports via the popular press and news stations. In our first year of data collection (2019), we enlisted 10 cats, and in our second year (2021), we enlisted 32 cats. In 2021, we added an online questionnaire (Article S1) that allowed us to record each cat’s sex, spay/neuter status, age, and personality score. The survey also allowed us to select participants with owned cats that spent unsupervised time outside (i.e., not in an enclosed outdoor space, or on a leash or tether) during both the day and night. Cats under 1 yr did not qualify because they were not fully grown, and their personalities could still be developing (Lowe & Bradshaw, 2001). Because the GPS units were attached to a collar and needed to be turned on by the owner before going outside (see details below), we also selected cats that had restricted outdoor access (i.e., did not use a cat door or other method of entering/exiting the house freely). All cats were selected from different owners and were classified as “owned indoor-outdoor” cats defined by (Crowley, Cecchetti & McDonald, 2019). Owned indoor-outdoor cats differ from free-ranging (e.g., cats that live outside entirely but have a colony caretaker that provides some shelter and food) and feral cats (no human influence) in that they have a central home they return to where they are fed and provided with shelter and owners have some control over cat movement (i.e., when a cat is put out) and reproduction (i.e., if a cat is fixed).

Owners were trained through written and video tutorials that we developed, including detailed instructions for attaching and operating the equipment. In 2019, meetings with owners occurred in-person, involving a single meeting for reviewing the informed consent form and teaching the owners how to use the equipment and fit the collar on their cat. Due to COVID-19 public health and University regulations restricting person-to-person contact, in 2021–22, we met with owners virtually to first go over the consent form and then again to observe them use the equipment and fit the collar around their cat. Before letting the cat outside with the equipment, owners provided their cat with a 1–2 d acclimation period where their cat wore the collar in the house so owners could observe if any abnormal behaviours occurred, such as excessive scratching at the collar, refusing to eat or play, or lethargy. If a cat showed any abnormal behaviour as reported by the owner, we removed them from the study. While owners had the equipment, we provided weekly check-up emails and video calls for technical assistance when required. As an incentive to participate, owners were given a map of their cat’s GPS tracks, a ‘highlight’ reel of their cat’s outdoor activity, a certificate of their cat’s personality score, and were entered into a draw for 1 of 4 $50 gift cards to a local pet store.

In total, we collected 63,474 GPS locations from 42 cats (18 males, 15 females, 9 unknown). Our sample size was limited by the number of camera devices, so no priori sample size calculations were performed. While we did not collect personal information about cats in 2019, we know through correspondence with owners that one was a male, and all were either neutered or spayed (hereafter, “desexed”). In 2021, out of 340 participants who responded to our survey, 121 owners had cats that qualified, and 34 were randomly selected to participate. One cat was withdrawn because the cat did not tolerate wearing the GPS, and one cat lost the tracker. This left 32 individuals (17 males, 15 females, all desexed), with the oldest cat being 14 yrs, and the youngest being 2.5 yrs (mean = 7 yrs, median = 6 yrs).

Filtering GPS data

We differentiated between “true” or “false” locations by filtering out points that were likely errors caused by interference between the GPS trackers and transmitting satellites (Moen, Pastor & Cohen, 1997; D’Eon et al., 2002). Specifically, to filter GPS data, we first determined what speed the cat would have traveled to reach each location based on the time and distance recorded between locations. Once we calculated a speed value for each GPS point in R v.4.1.2 (R Core Team, Vienna, Austria; Article S3), data were uploaded into ArcGIS Pro v.2.8 (Geographic Information System; ESRI, Redlands, CA., USA) for filtering. In ArcGIS, we removed highly improbable points based on date and time stamps (e.g., points that occurred at a researcher’s house or the University of Guelph campus). We then filtered out points over a speed threshold of 100 m/min (Recio & Seddon, 2013), classified as a trot (Smith, Chung & Zernicke, 1993), which was unlikely to be maintained by a cat for longer than 2 min. (Smith, Chung & Zernicke, 1993; Kim et al., 2014). Following this, we removed “spike locations”, which were characterized as points that could only occur in the unlikely event that a cat quickly ran to one location, then made a sharp turn to quickly return near the original location (Bjørneraas et al., 2010; Recio & Seddon, 2013). We used the same speed angle threshold values set by Recio & Seddon (2013) of 15 m/min with an outer turning angle between 165–180°. We then removed GPS points that occurred in-between points previously removed based on the above criteria. Finally, to further reduce bias in the movement and home range analyses, we removed all “stationary” points (points within 0 m of each other over successive locations) except the first point in a stationary series. We chose to do this because most stationary points occurred inside the owner’s home as a result of the owner not turning off the GPS immediately after the cat entered the house. Some stationary points may have been in the owner’s yard rather than the house, but we could not always distinguish these points from inside the house, so we chose the conservative approach to remove all stationary points. After filtering, there were 41,092 GPS points.

Home range size

We estimated home range size by using 100% minimum convex polygons (MCP) and 95% kernel density estimation (KDE). We analyzed both measures of home-range size to allow for comparison to other studies. Core home ranges, or 50% kernel density estimates (Samuel, Pierce & Garton, 1985; Heikkilä et al., 1996), are the most concentrated area of an individual’s activity (Burt, 1943) and exclude movement that occurs on the outskirts of the home range. Therefore, to estimate how far cats typically range, we focused on the 95% KDE and 100% MCP home ranges. To determine whether there were differences between the nocturnal and diurnal home range of cats, we split GPS points into day (06:00–17:59) and night (18:00–05:59). Because cats had sporadic schedules based on their owner’s behaviours and lifestyles, they would go in and out of the house relatively regularly when owners put them out. We based the specific timing of this split on when cats were inside for long periods of time. In the early morning, owners brought their cat inside to be fed before work, let them out again, brought them back in for a bit at the end of the workday, and put them out again sometime in the evening until the morning. These patterns were determined through conversations with owners. All home range analyses were carried out in R using the package ‘adehabitatHR’ (Calenge, 2006). We determined if sufficient GPS data were collected to accurately estimate home ranges by plotting the home range size vs. the number of GPS points. Through visual inspection for an asymptote (Harris et al., 1990), we determined that enough GPS points were collected for each cat (Article S4, Fig. S1).

Cat personality

To determine if a bold personality could impact home range size, we estimated the degree of neuroticism and assumed the inverse of this score was representative of boldness. Neurotic cats show traits of being shy and anxious, while cats that score low in neuroticism are the opposite —bold and secure (Litchfield et al., 2017). Neuroticism was measured by asking owners to complete a questionnaire (Article S5) based on an established personality test for domestic cats, called the ‘Feline Five’ (Litchfield et al., 2017). The Feline Five identified five personality types based on clusters of personality traits: (1) extraverted- traits such as inventive, persevering, and inquisitive; (2) agreeable- traits such as gentle, playful, affectionate, and friendly to people); (3) dominant- traits such as defiant, greedy, and unfriendly to other cats); (4) impulsive- traits such as distractible, erratic, and unpredictable; and (5) neurotic- traits such as shy, anxious, and insecure. The inter-item reliability of our personality measure was assessed using Cronbach’s alpha in R with the psych package (Revelle, 2024) and resulted in a coefficient of 0.92. Our questionnaire asked owners to evaluate 43 personality traits by rating how much they agreed or disagreed that their cat showed each trait on a 7-point Likert scale (Likert, 1932), with strongly disagree being the lowest score (1) and strongly agree as the highest score (7). Response scores were then summed and divided by the number of traits related to their respective personality type. Neuroticism was evaluated from the response scores of 13 traits (Litchfield et al., 2017) and we reported the standardized average score.

Road density

We estimated how road density and the presence or absence of major roads could impact home range size in ArcGIS by overlaying our GPS points of cat movement with road cover data from the Ontario Road Network (Ontario Ministry of Natural Resources and Forestry, 2019a). To estimate road density near each cat, we summed the length of roads sampled within a circular boundary centered on each individual’s mean coordinates. Because we wanted a boundary that captured how much road was available to each cat in a standardized way across cats, we used a radius of 66 m, which was based on an area the size of the median 95% KDE home range size. We examined if traffic acted as a barrier for cat movement by identifying roads as either “minor” or “major” based on Google Maps’ classification, which refers to traffic rates. We also “ground-truthed” (Wake & Hull, 1978) this classification by visiting the areas where all cats were tracked.

Land cover data

Land cover data were obtained from the Southern Ontario Land Resource Information System (SOLRIS) v.3 (resolution of +/−10 m and 30 m pixel resolution; Ontario Ministry of Natural Resources and Forestry, 2019b). We overlaid the land cover data with GPS locations from the cats that we tracked in ArcGIS. For our mapping purposes, we used the unmanipulated SOLRIS land cover categories (Fig. 1) for “impervious” (residential, industrial, commercial, and civic areas), “roads” (highways and roads), and “undifferentiated” (not mapped by the previous classes). We created a single land cover category for “greenspaces”, consisting of marsh lands, treed swamps, forests, and urban recreational lands (i.e., golf courses, and parks), and a category for “agriculture”, which represented the sum of tilled land and cultivated tree plantations. Undifferentiated areas were comprised of variable land cover types ranging from agricultural features to transportation rights-of-ways; therefore, any potential relationship with this land cover could not be readily interpreted. Thus, we removed all points that fell within this category (n = 27,469 or 3% of all points, n = 901,157). We also removed points that fell within water because this habitat type was only present in the buffer of three cats which made up <0.5% (n = 1,772) of the available land and was not used by any of the cats.

Statistical analysis

Habitat selection

To determine if cats showed a selection for land cover types, we used resource selection functions (RSF) to compare the land covers cats used with the land covers available to them (Manly et al., 2002). While second-order selection represents an animal’s selection of home range within a geographical area, and third-order selection describes the space within the home range, selection functions can be used to examine habitat availability at scales that fit between these orders (Northrup et al., 2021). We performed two RSF analyses at the third-order to compensate for a cat’s habitat selection being influenced by their owner. Therefore, we compared land cover use in a boundary around each owner’s house and within the home range (Johnson, 1980; van Moorter et al., 2016), detailed below.

We derived both RSFs by comparing the “used” GPS points to “available” points that were randomly generated anywhere within the set boundary at a ratio of 10:1 (10 random points per 1 used point; (Fieberg et al., 2021). To evaluate land cover used close to the owner’s house, we sampled the used and available points within a circular radius of 300 m, centered around each owner’s house. We used a boundary size of 300 m to stay consistent with previous research (Pirie, Thomas & Fellowes, 2022), which examined cat habitat selection within a site equivalent to a recommended buffer zone size of ∼300 m around a natural area (328 m, Thomas, Baker & Fellowes, 2014; 360 m, Lilith, Calver & Garkaklis, 2008). We removed one cat from this analysis because we did not have the exact address of the owner. For the second analysis, we sampled used and available locations within each cat’s 100% MCP home range.

We estimated the likelihood that a habitat type was selected by an individual by using binomial generalized linear mixed models (GLMM) with a logit link function. We constructed separate GLMMs for each habitat association close to the owner’s house and within the entire home range. In both analyses, we included a binary response variable (1 = used, 0 = available) to indicate if the cat used the land cover type or if it was available, where predictor variables included each land cover type at the point and whether it was used or not or (1 = within the habitat type, 0 = outside of the habitat type). To account for multiple locations occurring for individuals, cat ID was used as a random effect. The selection coefficient (β) from the RSF model indicates selection strength quantitatively for each land cover type with a positive or negative value reflecting a selection or not (Boyce et al., 2016; Avgar et al., 2017), respectively, and quantitatively with the β approaching 0 indicating a weaker selection (Avgar et al., 2017).

Home range size

Our analysis of 42 individuals captured a high degree of variation in both the 95% KDE and 100% MCP home ranges (Fig. 3; Article S4, Appendix S4). The median 100% MCP was 4.40 ha with a range of 0.34–38.45 ha. While individual 95% KDE home range sizes were ∼4 times smaller, they showed similar variation between individuals (Fig. 4; Article S4, Appendix S4). The male and female 100% MCP home ranges had similar variation (Brown-Forsythe test, F = 0.28, p = 0.60; Fig. 4), with the median male 100% MCP home range of 6.15 ha (range: 0.34–38.45 ha) and a median female 100% MCP home range of 4.40 ha (range: 0.52–30.16 ha). We did not find evidence to suggest cats had larger home ranges at night than during the day (no support for hypothesis 2; Wilcox signed rank test, W = 410, p = 0.61). During the day, the median 100% MCP was 2.63 ha (range: 0.15–34.45 ha), while, at the night, the median 100% MCP was 3.63 ha (range: 0.14–22.88 ha; Fig. 5). The diurnal and nocturnal 95% KDE home range size showed similar patterns (Fig. 5; Article S4, Appendix S4).

With respect to intrinsic predictor variables from 29 cats, the top ranked model was the null model and the second-ranked model (ΔAICc = 0.144) included age. Because the confidence intervals for sex (β = −0.24, 85% CI: −0.84, 0.37), age (β = 0.05, 85% CI: −0.02, 0.13; Table 1), and neuroticism (β = 0.002, 85% CI: −0.01, 0.02) did overlap with zero (no support for hypotheses 1, 3, or 5), we did not carry forward these predictors in the subsequent evaluation of extrinsic models. When considering which extrinsic variables could influence the size of the home range, there were four candidate models within 2 ΔAICc of the top model (Table 2). Based on the model averaged coefficient estimates and confidence intervals, 100% MCP home range size was positively related to road density (Table 3), suggesting traffic did not act as a barrier for cats (no support for hypothesis 4). The analysis of 95% KDE followed similar patterns to the 100% MCP models, but home range size was positively related to both road density and time of year (Article S4, Appendix S4). Data are available on Figshare: https://doi.org/10.6084/m9.figshare.25326322.v2.

Habitat selection

Based on the random points generated within individual 100% MCP home ranges from 42 cats, the most abundant available land cover types were impervious surfaces (66%), roads (26%), greenspaces (6%), and least abundant was agricultural land (2%). Within the 100% MCP home ranges, greenspaces were predominately comprised of recreational parks (89%), and then various natural habitats (thicket swamps (4%), treed swamps (3%), deciduous forests (3%), and marshlands (1%)). Results from the GLMM provided some support for hypotheses 6 and 7, as cats selected to use impervious surfaces (73% of used locations) but avoided roads (22% of used locations) and greenspaces (4% of used locations), in proportion to their availability (Table 4, Fig. 6). There was also evidence that they avoided agricultural land (1% of land used; Table 4, Fig. 6). Although cats showed an avoidance of roads, used locations showed that 90% (n = 38) of cats utilized a road at some point during the study. The model results for habitat selection for a buffer close to the owner’s home were similar to the 100% MCP results (Fig. 6; Article S4, Appendix S5). Habitat data are available on Figshare: https://doi.org/10.6084/m9.figshare.25154555.v2.

Management Implications and Recommendations

Based on our findings, we provide four main recommendations for interest groups to consider when managing outdoor cats. First, despite a general avoidance of roads, 90% of cats in this study crossed roads or walked alongside them placing them at risk of blunt force trauma or death from vehicle collisions. Findings such as these may be valuable for interested parties such as veterinarians or shelter workers to present and discuss with cat owners to help them assess the risk for allowing their cat(s) to wander outside unsupervised. Second, consistent with Clyde et al. (2022), our finding of a close association with residential and urban land covers indicate that native wildlife species found close to buildings are at the greatest risk of predation, and may have implications for backyard bird feeders (Dunn & Tessaglia, 1994) or for species that can take advantage of the novel ecosystems created by humans (e.g., song sparrows (Melospiza melodia), blue jays (Cyanocitta cristata), and cedar waxwings (Bombycilla cedrorum)). Third, our results suggest that cat curfews are unlikely to reduce the spread of diseases carried by cats and the magnitude of risks that cats face while outside, as cats roamed the same distances during the two time periods we examined. Lastly, based on our habitat selection analysis, buffer zones around greenspaces may be of limited use for eliminating predation rates of wildlife by owned cats in urban habitats. That said, if the primary goal of a community is to prevent wildlife predation in greenspaces by cats, then buffer zones could be an option. To estimate a buffer zone based on our results, we added 20% safety margin (Lilith, Calver & Garkaklis, 2008) to the largest 100% MCP home range (38.45 ha, which is equivalent to a diameter of 700 m), resulting in an 840 m buffer. This conservative approach would encompass even occasional cat forays into natural habitats. Regardless, results of actual predation events from the Catcams (the cameras attached alongside the GPS units) that will arise from this study should provide a better understanding of where and when predation is greatest, information that will complement data on cat habitat use; therefore, providing a more comprehensive picture of predation risk across the urban landscape.

Our findings highlight that the movement of outdoor domestic cats is highly variable among individuals. This finding is consistent with other studies that have shown high individual variation in hunting behaviour (Morgan et al., 2009; Tschanz et al., 2011; Dickman & Newsome, 2015; Cecchetti, Crowley & McDonald, 2020) and risk-taking behaviours (Loyd et al., 2013; Bruce et al., 2019), making it hard to develop a one-size-fits-all approach to cat management. While a buffer size of 840 m may protect native wildlife from cat depredation in our study area, this could be an overestimate (Lilith, Calver & Garkaklis, 2008; Thomas, Baker & Fellowes, 2014; Hanmer, Thomas & Fellowes, 2017) or underestimate (Mesters, Seddon & van Heezik, 2010) in other locations. Given the complexity of urban environments and our findings showing how cat movement was related to surrounding habitat, it is hard to develop a one-size fits all buffer size that would be appropriate in all cities. Additional research on cat movement may help reduce some of these complexities and provide a way to better assess how to adjust buffer size based on surrounding habitat, but first it is valuable to consider if such research is necessary to meet cat management or wildlife conservation objectives. For municipalities or non-governmental organizations that are considering implementing management strategies that incorporate buffers they would likely benefit from conducting locally or regionally based research to collect relevant movement data from local cats. This could be done with a relatively small sample size of 20–30 individuals because, as our research has shown, this is enough to capture variation in home range sizes.

Based on the precautionary principle, which advocates for taking preventative action when there is uncertainty (Kriebel et al., 2001), and the high variability of cat movement, restricting cats indoors would be the main way to fully reduce their environmental impacts. However, this would ignore the complexity of issues surrounding outdoor domestic cats and society’s perceptions of cats, such that this management strategy would be impractical and likely infeasible. For example, the primary reason owners keep their cats inside is the concern for their cat’s safety or wellbeing (MacDonald, Milfont & Gavin, 2015; Tan et al., 2021; van Eeden et al., 2021) and not because cats hunt wild animals (Wald, Jacobson & Levy, 2013; van Eeden et al., 2021), suggesting that focusing on environmental impacts is not an effective communication strategy (Gramza et al., 2016; Crowley, Cecchetti & McDonald, 2020b; van Eeden et al., 2021). We encourage future research to continue to broaden understanding how cats behave outside to develop the best management practices and provide stakeholders such as shelter workers and veterinarians with information that can help owners make informed decisions about their cat’s outdoor roaming activities.