Is time partitioning the currency of coexistence for a grassland canid community?

Habitat and diet preferences are often considered major aspects of niches differentiation among species; however, partitioning of habitat and resource use temporally is often overlooked in modeling coexistence. The plasticity of temporal activity patterns of individuals may influence the species’ response to selective forces and long-term persistence. Temporal avoidance may be a mechanism by which subordinate species can reduce the likelihood of direct competition with dominant sympatric species. Here, we examine temporal activity patterns of three canid species (swift fox Vulpes velox , red fox Vulpes vulpes and coyote Canis latrans ) to determine how temporal activity patterns and activity overlap varies among species. We found that all species presented seasonal differences in activity patterns. When activity patterns were compared among species, the estimates of activity overlap in the spring season (i.e. breeding and pup-rearing periods) were higher than the activity patterns in the fall (i.e. juvenile dispersal and pair-formation periods); however, activity pattern overlap among species was significant only during the fall. Overall, these patterns revealed a close temporal overlap between swift fox (subordinate species) and both red fox and coyote (dominant species), which provides insight into conditions under which time partitioning may not be as clear as predicted, and other mechanisms may facilitate species coexistence. Considering swift fox population decline and distribution contraction across the North American grasslands, investigating temporal activity patterns of the canid species may reveal the implications of altering such patterns for individual animals, populations and ecosystems.


Introduction
Understanding the mechanisms driving patterns of distribution, diversity and abundance of species in ecological communities is crucial for wildlife conservation (Farris et al. 2015).Community structure is shaped by multiple spatiotemporal interactions within and among species.Classical ecological niche theory (Hutchinson Is time partitioning the currency of coexistence for a grassland canid community?Lucia Corral, Teresa J. Frink and Joseph J. Fontaine L. Corral (https://orcid.org/0000-0002-5153-2678)✉ (luciacorral@gmail.com)and J. J. Fontaine, Nebraska Cooperative Fish and Wildlife Research Unit and School of Natural Resources, Univ. of Nebraska-Lincoln, Lincoln, NE, USA.-T.J. Frink, Chadron State College, Chadron, NE, USA. 1957, 1959, MacArthur and Levins 1967) proposes that the coexistence of species that fill similar ecological roles is facilitated by differences in resource use involving the segregation of habitat, food or time (Schoener 1974, Kronfeld-Schor andDayan 2003).Habitat and food are most often considered the primary resources for niche differentiation (Schoener 1974); however, partitioning the temporal niche dimension as a mechanism enabling coexistence is equally important (Pianka 1973, Richards 2002, Kronfeld-Schor and Dayan 2003, Farris et al. 2015, Dröge et al. 2017).Temporal activity patterns of individuals, which are affected by ecological and physiological costs and constraints, may influence the plasticity of a species' response to selective forces and, therefore, the potential persistence of a species through time (Halle andStenseth 2000, Kronfeld-Schor andDayan 2003).Investigating temporal activity patterns can aid in understanding behavioral and ecological components of the life history of a species; for example, by giving insight into the complex balance of risk-avoidance and energetic needs (Rowcliffe et al. 2014).
In some guild communities, such as among canids, there is evidence that top predators kill and harass smaller predators (Macdonald and Sillero-Zubiri 2004a, b), and such interactions (i.e.interference competition) affect the distribution and population dynamics of the smaller species (Voigt andEarle 1983, Swanson et al. 2016).Temporal avoidance may be a mechanism by which subordinate species can reduce the likelihood of direct interference competition with dominant sympatric species and promote coexistence.In the Canidae family, where interference competition appears critical, larger species such as coyotes Canis latrans, can often affect smaller species, such as swift fox Vulpes velox and kit fox Vulpes macrotis, by killing or displacing foxes (Harrison et al. 1989, Cypher and Spencer 1998, Ralls and White 1995).
Swift foxes and coyotes in North American grasslands are believed to have considerable overlap in habitat use, home range requirements, food habits and reproductive timing (Kamler et al. 2003, 2007, Corral et al. 2021).As the largest canid in the grasslands, coyotes are dominant to the swift fox and are often cited as an important source of mortality (Covell 1992, Sovada et al. 1998, Schauster et al. 2002, Karki et al. 2007).Similarly, the red fox Vulpes vulpes is considered to be a barrier preventing swift fox populations from expanding into unoccupied but suitable areas (Sovada et al. 1998).In response to habitat loss, persecution and interference competition, swift foxes have been displaced across most of their historical range (Allardyce and Sovada 2003, Sovada et al. 2009), and are currently listed as a threatened or endangered species by half of the states in their historic range.Coyote and red fox, on the contrary, have increased both in abundance and range throughout North America (Hill et al. 1987, Lovell et al. 1998, Gompper 2002, Prange and Gehrt 2007).Declines in distribution or abundance of a specialist species due to changes in the habitat in which they specialize (e.g.grassland obligates; Sampson and Knopf 1994) may increase the importance of temporal partitioning, particularly if the ecosystem changes favor increases in dominant habitat generalists (Benedict et al. 1996).
Here, we examined the temporal activity patterns of three canid species (swift fox, red fox and coyote) to determine how activity varies among species and quantify activity level overlaps.We focused on three main questions: 1) Is there temporal segregation among species?2) Is the degree of temporal segregation predicted by body size, a proxy for dominance among canid species?And 3) Is temporal segregation consistent through time, or is it determined by trade-offs driven by predictable shifts in the species' life history?
Ultimately, we aimed to assess temporal separation as a mechanism that favors the coexistence of three canid species that compete in areas of sympatry (Covell 1992, Carbyn et al. 1994, Ralls and White 1995, Sovada et al. 1998, Kitchen et al. 1999, Schauster et al. 2002, Andersen et al. 2003, Thompson and Gese 2007), and the implications of landscape changes in altering such patterns.

Study area and species
Our study area encompassed approximately 68 605 km 2 of western Nebraska, USA.(Supporting information).The landcover was primarily native shortgrass and mixed grass rangeland, with some areas converted to corn, wheat and sugar beets (Bishop et al. 2011, Schneider et al. 2011, Corral et al. 2021).Additional patches of other native habitats, such as woodlands and wetlands, are scattered throughout the study area.The region presents a relatively diverse topography, including several areas of rocky escarpments and a great variety of soil types, ranging from sands to heavy clay.The climate is semi-arid, characterized by low humidity, moderate to high winds, and a large daily and seasonal range in temperature (Chapman et al. 2001, Schneider et al. 2011).Annual precipitation ranges from 300 to 430 mm, with average wind speeds ranging from 14 to 24 km h −1 , average winter temperatures ranging from −7° to −4°C, and average summer temperatures ranging from 22° to 26°C (Anderson 1999, Schneider et al. 2011).
The swift fox is endemic, restricted to the shortgrass and mixed-grass prairies, and is the smallest canid species in North America (average weight of 2.4 kg; Moehrenschlager and Sovada 2004).The red fox is found in diverse habitats such as shrubland, bushland, forested areas, grasslands, mixed agricultural habitats and on the margins of some urban areas, and is an intermediate size (average weight of red fox 5.8 kg; Macdonald and Reynolds 2004).Lastly, coyotes are the largest (average weight of 10.8 kg) and the most widespread canid, living in almost all available habitats throughout the study area (Gese andBekoff 2004, Sillero-Zubiri et al. 2004).Differences in canid body masses are correlated with differences in prey body mass -e.g.larger species specialize in larger prey -and consequently, the use of space can differ among species (Rosenzweig 1966, Carbone and Gitlleman 2002, Radloff and Du Toit 2004).Nonetheless, in the grasslands, the three canid species use similar habitats, have overlapping prey use (swift foxes and coyotes: Kitchen et al. 1999, Kamler et al. 2007;red foxes andcoyotes: Fuller andHarrison 2006, Mueller et al. 2018), and all species are thought to be primarily active at night (Sillero-Zubiri et al. 2004).Coyotes exhibit interference and exploitation competition with both fox species, especially swift fox, through intraguild predation (Kamler et al. 2003, Sillero-Zubiri et al. 2004, Nelson et al. 2007).Therefore, variation in the use of food resources and space may not be sufficient to describe and explain the canid community structure, where activity patterns may play an essential role in determining interspecific relationships (Jacomo et al. 2004).

Data collection
We investigated activity patterns and temporal overlap among canid species through camera-trap records obtained during surveys conducted from March to May (spring season) and from September to November (fall season) of 2014 and 2015, and in the fall of 2016.To optimize the detection of swift fox, the species expected to be the most difficult to detect, we selected survey sites across different land covers known to be important to swift fox (see below for details; Finley 1999, Finley et al. 2005).To increase detection rates and geographic extent, while reducing issues of pseudoreplication, we divided the study area into a grid of 31 km 2 grid cell (total number of grid cells = 2331), a resolution meant to approximate a swift fox home range (Hines and Case 1991, Finley 1999, Finley et al. 2005).We classified each grid cell by the percentage of potentially suitable habitat for swift fox based on an a priori habitat suitability map and the slope layer.A grid was defined as 'suitable' (number of suitable grid cells = 1737) if it was composed of > 25% suitable land cover (i.e.short-and mix-grass prairie) and > 45% suitable slope (i.e.areas that present < 10% of slope), characteristics that reliably predicted occupancy and detection of swift foxes (Finley et al. 2005, Martin et al. 2007, Knox and Grenier 2011).
We used the spatially balanced points tool, which uses a reverse randomized quadrant-recursive raster algorithm (RRQRR;ArcGIS ver. 10.3.1 (ESRI 1995-2015)), to sample the grid randomly.The RRQRR algorithm selects from all available grid cells (n = 2331), taking into account the potential spatial pattern of a population, and optimizes sampling based on the probability of observing a target species in a specific point given the percentage of suitable habitat (Stevens andOlsen 2004, Theobald andNorman 2006).We established our survey sites at the selected grid cells depending on landowner permission and terrain constraints.
At each grid cell selected for sampling (n = 207), hereafter referred to as 'site', we placed an average of 4.24 (SD = 2.73) trail cameras (Bushnell Trophy Cam HD and Moultrie M-880 models) with olfactory lures on existing trails (e.g.cow trails, unpaved roads) or fence lines (Knox and Grenier 2011).Camera-trap stations (n = 902) were spaced a minimum of 1.6 km apart to maximize detection rates within sites (Bushnell cameras: optical field of view = 45°, approximate detection range = 12 m, response time = 0.6 s; Moultrie cameras: optical field of view = 50°, approximate detection range = 12 m, response time = 0.8 s).A camera was hung on a post 40 cm above the ground at each camera-trap station, and the location was recorded using a hand-held GPS (Garmin eTrex 10).We set a wooden stake 3 m in front of each camera with 40 cm exposed above the ground, which served as a base for the lure, a focal point for the camera and a metric for estimating animal body size (Hegglin et al. 2004).The lure consisted of approximately 15 ml of a skunk-based attractant produced by heating 385 ml of petroleum jelly to liquid form, adding 15 ml of skunk essence (F&T Fur Harvester's Trading Post, Alpena, MI) and allowing the lure to solidify.The distance between cameras was chosen to optimize scent attraction based on volatilization rates of fatty acid to maximize the detection rates at individual camera-trap station (Roughton and Sweeny 1982, Kahn et al. 1997, Harrison et al. 2002, Sargeant et al. 2003).Cameras were set up to take bursts of 3 photographs no less than 5 s apart each time motion and heat signature were detected.We left cameras-traps running for a minimum of 10 consecutive nights (mean = 13.67 nights) to balance the trade-off between detection probability and sampling time.

Data analysis
We manually processed all images (n = 6 555 920) using Timelapse Image Analyser software (Greenberg andGodin 2012, 2015).We eliminated all dark and corrupted images (n = 51 550) and then identified all pictures of canids to species.The resulting data for each camera was associated with a unique GPS location and saved as minute-by-minute detection histories (i.e. a detection was recorded when at least one individual of the target species was photographed during each minute of the survey).Under the assumption that activity patterns were similar across the years of our surveys and due to the small sample size for rare species, we pooled all records across years (2014, 2015 and 2016).We treated each picture of a canid as a separated data point, in some cases including multiple pictures of the same individual in the analysis (Carver et al. 2011).We assumed that the probability of individual detection during a single minute approximates activity at the population level and reflects individual-level activity trade-offs, including competition and predation risk.
Because daylight length varies seasonally and our target species are described as predominantly nocturnal, we adjusted each record's 'clock time' to the specific sunrise and sunset of that date at that location.We then converted to a day of 12-h length with sunrise at 06:00 h and sunset at 18:00 h, which allowed us to standardize temporal and geographical variation in daylight (Carver et al. 2011, Nouvellet et al. 2012).Sunrise and sunset were estimated based on date, location and the algorithms provided by the National Oceanic and Atmospheric Administration (NOAA) Page 4 of 13 using the 'sun-methods' function in R-package 'maptools' (Bivand andLewin-Koh 2013, 2018).
We determined the temporal activity patterns of our target species and estimated the overlap of activity patterns between the species using a two-step procedure.First, kernel density curves were fitted to the data using a non-parametric von Mises kernel density function corresponding to a circular distribution (Ridout and Linkie 2009, Rowcliffe 2016, Meredith and Ridout 2018).Second, the curves were compared to each other based on the degree of overlap in the area lying under the density curve of individual species (i.e.coefficient of overlapping; Weitzman 1970).Specifically, the coefficient of overlapping (∆) is defined as the area under the curves formed by the minimum of the two density functions at each point in time.The value of ∆ lies between 0 and 1, with ∆ = 0 if there is no overlap and ∆ = 1 if there is complete overlap (Ridout and Linkie 2009, Linkie and Ridout 2011, Meredith and Ridout 2018).We used two estimators of ∆, labeled ∆ 1 and ∆ 4 (for consistency with Ridout and Linkie 2009) with equivalent mathematical expressions but adjusted to different sample sizes such that ∆ 1 performs better for smaller samples and ∆ 4 performs better for larger samples (Ridout and Linkie 2009, Meredith andRidout 2018).We used ∆ 1 for samples < 50 records and ∆ 4 for samples > 75 records (the estimator was chosen based on the size of the smaller of the two samples) and used smoothing parameters 0.8 and 1.0 to estimate ∆, respectively (Ridout and Linkie 2009, Meredith and Ridout 2018).We calculated 95% confidence intervals of each overlap index using smoothed bootstrap with 10 000 resamples (Azevedo et al. 2018, Meredith andRidout 2018).
Since the coefficient of overlap is merely descriptive, we used the function 'compareCkern' in the R-package 'activity' (Rowcliffe 2016) to test the probability that the two sets of circular observations come from the same distribution.'CompareCkern' uses a randomization test that calculates an overlap index for the observed data samples.'CompareCkern' then generates a null distribution of overlap indices using data sampled randomly with replacement from the combined data and uses the randomized distribution to estimate the probability that the observed overlap is given by chance (Ridout and Linkie 2009, Rowcliffe 2016).Additionally, we computed a Wald test for each pair of activity level estimates to compare activity patterns.All analyses were performed in R ver.3.4.3(<www.rproject.org>)using the 'overlap' and 'activity' R-packages (Rowcliffe 2016, Meredith andRidout 2018).
Finally, we investigated the potential relationship between our target species' coefficient of activity overlap and the detection density (i.e.number of detections per unit of area) for coyotes.As measured here, detection density is not a measure of coyote population density, but rather a proxy of encounter risk between swift fox and coyote, reflecting both the number of coyotes and their relative activity rate within the sample area.We expected the coefficient of overlap between subordinate species and coyotes to be proportionally related to coyote detection density, assuming that subordinate species can minimize competition by inhabiting areas with either fewer dominant competitors or dominant competitors that are less active independent of the degree of temporal overlap between the species (Hayward and Slotow 2009).As our camera-trap locations were clustered within sites, first, we calculated coyote detection density using the number of detections within a circular neighborhood surrounding the site (area = 124 km 2 ).For this analysis, we consider as a single record consecutive photos that were taken < 60 s apart (i.e. the presence of the same individual for several consecutive minutes represented a single detection, and the next individual detection would begin > 60 s after the last individual was out of the camera's field of view).Second, we estimated each species' activity pattern and the coefficient of overlap between species separately for each site cluster.Third, we fitted linear regression to examine the coefficient of overlap as a function of coyote detection density.We analyzed the data on a log scale (natural logarithm).

Results
Across all years, we obtained 929 633 records with a total of 23 136 pictures of canids (Supporting information) -2298 pictures of swift fox (0.10 detections/trap night), 1306 pictures of red fox (0.06 detections/trap night) and 19 532 of coyote (0.84 detections/trap night).Out of the 902 camera-trap stations, swift foxes were recorded at 63 camera-trap stations, red fox at 43 and coyote at 638.We attained 4371 activity records (by minute) of the three canid species, including swift fox (n = 442), red fox (n = 226) and coyote (n = 3703).The three canid species were detected primarily between midnight (24:00 h) and noon (12:00 h) and considerably fewer records were obtained during the spring (n = 980) than the fall (n = 3391; Fig. 1).

Overall temporal activity patterns
Swift fox, red fox and coyote concentrated their daily activity between midnight and a few hours after sunrise, but all the species presented some difference in their activity cycles.Swift foxes showed higher activity levels between midnight and sunrise and were more active during the morning hours than red foxes and coyotes.Red fox presented two distinct peaks of activity, one between midnight and sunrise (02:30 h) and another right before noon (11:00 h).Coyote was active mainly between midnight and noon (Fig. 2).The comparison of activity patterns between canid species showed a mean coefficient of activity overlap (∆) of 0.73 (SE = 0.048).The highest activity overlap was observed between coyote and swift fox (∆ = 0.81, p < 0.001; Fig. 3a), followed by coyote and red fox (∆ = 0.73, p < 0.001; Fig. 3c) and the lowest activity overlap between red fox and swift fox (∆ = 0.65, p < 0.001; Fig. 3b).
When we examined the relationship between the differences in body weight of the canid species and the degree of temporal segregation (i.e.equal to 1 -coefficient of activity overlap) between the species, we found the variables seemed to present a negative linear relationship, as the difference in body weight increases the degree of temporal segregation decreased; however, this observation is inconclusive because the analysis was limited to only three aggregated data points (i.e.averages of body weight and coefficients of activity overlap; Supporting information).

Seasonal activity patterns overlap
The overlap of seasonal activity patterns ranged from ∆ = 0.56 to ∆ = 0.84 and was statistically different for all the target species (Table 1, Fig. 4).Swift fox showed the lowest overlap in activity patterns between spring and fall (∆ = 0.56, p < 0.005).During the spring, swift fox presented a distinct initial peak of activity approximately four hours after midnight (03:50 h), and activity declined gradually afterward until noon.During the fall, swift foxes did not seem to present an initial peak of activity but a steady increase of activity after sunset, with maximum activity between 01:00 and 02:00 h, followed by relatively constant activity until a decline in the afternoon.Minimum activity occurred between 18:00 and 19:00 h (Fig. 4a).Red foxes showed a higher seasonal activity overlap (∆ = 0.62, p = 0.005) than swift foxes but lower than coyotes.Red foxes exhibited a marked peak of activity around sunrise (05:00-07:00 h) during the spring but two distinct peaks of activity during the fall, one between midnight and sunrise (02:30 h) and another right before noon (11:00 h; Fig. 4b).Coyotes had the highest seasonal overlap of activity patterns between spring and fall (∆ = 0.84, p < 0.005), characterized by a high peak after midnight (spring: 03:00-03:30 h; fall: 01:00-02:00 h) and a steep decline before noon (Fig. 4c).Compared with either red foxes or coyotes, swift foxes exhibited different overlap coefficients each season (Table 2).No differences were found between activity pattern overlap in the spring, but the overlap estimates were significantly different in the fall (Table 2, 3).However, the spring's overlap estimate confidence intervals are much wider due to the smaller sample size.Swift foxes have the highest overlap with coyotes (∆ = 0.88) in the spring and the lowest overlap with red foxes (∆ = 0.60) in the fall.Although the seasonal activity patterns are different from those of swift fox, red fox presented similar values of coefficient of overlap with coyotes in both seasons (spring: ∆ = 0.89; fall: ∆ = 0.70), but only the fall overlap coefficient was statistically significant (Table 3, Fig. 5).

Coyote detection density and activity coefficient of overlap
Coyote detection density ranged from 0.12 to 4.06 detections km −2 in locations where swift fox and coyote were detected together and 0.02-2.12detections km −2 in sites where red fox and coyote were detected.Where they cooccurred, swift fox and coyote activity coefficients of overlap ranged from 0.29 to 0.75, and red fox and coyote coefficients of overlap ranged 0.14-0.60.Coyote detection density did not predict coefficient of overlap for swift fox (Supporting information; F (1,14) = 0.130, p = 0.724, R 2 = 0.009; Supporting information) or red fox (F (1,14) = 0.3543, p = 0.561, R 2 = 0.025;   Supporting information); however, only 16 sites for each species pairing had the minimum number of records to estimates overlap, substantially limiting our statistical power.

Discussion
We examined temporal partitioning as a mechanism to facilitate sympatry among canids in the plains of western Nebraska,    but contrary to classical ecological niche theory (Hutchinson 1957, 1959, MacArthur and Levins 1967), we found no temporal differentiation among species despite significant spatial (Corral et al. 2021) and ecological (White et al. 1995, Kitchen et al. 1999, Sovada et al. 2001, Moehrenschlager and Sovada 2004, Moehrenschlager et al. 2004, Kamler et al. 2007) overlap.The lack of temporal partitioning is surprising because carnivores tend to exhibit substantial temporal differentiation (Di Bitetti et al. 2009, Wang and Fisher 2012, Monterroso et al. 2014), in large part due to the significant dangers to subordinate species of competitive encounters (Palomares and Caro 1999).Although our results are inconclusive, body weight was associated with the degree of temporal separation (Macdonald andSillero-Zubiri 2004a, b, Sunarto et al. 2015) as red fox and swift fox (mean weight difference of 3.4 kg) showed the highest degree of temporal partitioning (35%), followed by red fox and coyote (27%; mean weight difference of 5.0 kg), and, lastly, swift fox and coyote (18%; mean weight difference of 8.4 kg).Our findings align with evidence that dominant species are more likely to harass and potentially kill only those subordinate species that are sufficiently smaller than themselves to minimize the risk of injury (Sargeant et al. 1987, Peterson 1995, Ralls and White 1995, Tannerfeldt et al. 2003), leading to greater partitioning among species of similar size.In our study area, however, red foxes are rare and relate to different habitat conditions than swift foxes (Corral et al. 2021), limiting the need for temporal partitioning, suggesting other mechanisms (e.g.human activity) may mediate differences in activity patterns between the species.In contrast, coyotes and swift foxes overlap in habitat (Corral et al. 2021), and depredation by coyotes is among the leading causes of swift fox mortality (Covell 1992, Carbyn et al. 1994, Sovada et al. 1998, Kitchen et al. 1999, Schauster et al. 2002, Andersen et al. 2003, Thompson and Gese 2007); yet, even at sites where the detection densities of coyote, a proxy interference competition risk, were relatively high, swift fox failed to show temporal avoidance of the times when coyotes were most active.
The high degree of spatial overlap (Corral et al. 2021) and lack of temporal partitioning among canid species may suggest that the density of the canid community in western Nebraska is below the threshold at which spatial or temporal avoidance is necessary for carnivore coexistence (Holt and Polis 1997, Palomares and Caro 1999, Chesson 2000, Caro and Stoner 2003).Although we did not estimate the abundance or density of any canid species, detection rates for all three species were low, as were occupancy rates of swift foxes and red foxes (Corral et al. 2021).An extreme drought in 2012-2013 affected prey populations (Laskowski et al. 2017), with likely cascading effects on canid populations (Ralls and White 1995).Prey populations can respond rapidly following drought (Bradley et al. 2006), and the consequences of drought can be more extreme for carnivores than herbivores (Prugh et al. 2018), creating a situation where competition may be reduced as the recovery of predator populations lags behind the recovery of their prey.The apparent lack of spatial or temporal partitioning within the canid community of western Nebraska may simply reflect the complexity of top-down and bottom-up processes in shaping predator community dynamics (Elmhagen andRushton 2007, Hayward andSlotow 2009), whereby resource partitioning may only be advantageous when costs from competition are high.The lack of temporal partitioning by swift foxes may indicate that the threshold of risk during our study was not significant enough to offset the cost to foraging, a conclusion that is supported by the lack of relationship between coyote detection density and swift fox activity.Assuming populations of all canid species were depressed, recovering canid populations should lead to increased partitioning by swift fox (Chesson 2000).However, even in communities where coyote and swift fox populations are thought to be at carrying capacity, swift foxes fail to demonstrate temporal partitioning (Kitchen et al. 1999), suggesting that the costs of partitioning are too high (i.e.reduced food acquisition).As with other carnivore communities (Dröge et al. 2017), swift foxes and coyotes may only coexist intermittently as populations of the dominant predator (Jachowski et al. 2020) or resource availability (Palomares and Caro 1999, Caro and Stoner 2003, Valeix et al. 2007) fluctuate across space and time (Thompson and Gese 2012).
We failed to find support for temporal partitioning by swift foxes, but we did find substantial within species temporal niche breadth.Swift foxes' activity has been described as predominately nocturnal and crepuscular (Andelt 1995,  (Cavallini and Lovari 1991, Doncater and Macdonald 1997, Hayward and Slotow 2009, Monterroso et al. 2014).Within carnivore communities, the high costs of interference competition often lead to spatial and temporal differentiation by subordinate species at relatively large spatial and temporal scales (i.e.proactive), but the risk varies across multiple spatial and temporal scales (Lima and Dill 1990), and individuals respond dynamically to risk at appropriate scales (Messinger et al. 2019), thus partitioning can occur at fine scales as well (i.e.reactive; Dröge et al. 2017, Creel 2018, Ferreiro-Arias 2021, Kautz et al. 2021).That swift foxes do not demonstrate proactive spatial or temporal partitioning may suggest that swift foxes are responding reactively at much smaller spatial and temporal scales to the presence of competitors (Broekhuis et al. 2013, Creel et al. 2018), resulting in a broad temporal niche breadth, similar to changes in diet breadth (Hayward andKerley 2008, Monterroso et al. 2020).Indeed, based on a post hoc analysis of 'time since detection,' at locations where swift foxes and coyotes cooccurred, when a swift fox was detected first (n = 10) a coyote was detected on average 8.45 (SE 1.00) hours later and when a coyote was detected first (n = 6) a swift fox was detected on average 3.42 (SE 1.01) hours later (Supporting information).Although, samples were too small to be definitive, the pattern suggests subordinate species maybe responding reactively rather than proactively to the temporal and spatial risk of interference competition.
Swift fox ecology and the ecology of the canid community of western Nebraska more generally would support such an assertion.All three species occur at relatively low densities, occupy large home ranges and are very mobile, so although there is considerable spatial and temporal overlap at relatively large scales, at finer scales, the risk of interspecific encounters may be extremely low and unpredictable.Unpredictability favors reactive rather than proactive responses (Creel et al. 2018), but even though the presence of canid competitors may be unpredictable, it is not unannounced.Canids actively communicate using olfactory and audio cues that can be used to assess risk and invoke reactive responses (Luttbeg et al. 2020, Edwards et al. 2021).Indeed, the tendency for swift foxes to associate with flat, sparsely vegetated, open spaces may facilitate risk assessment in their environment as olfactory, auditory and  visual cues may be less obstructed (Kitchen et al. 1999).The unpredictable but easily assessed risk from competitors favors a plastic response (Turcotte and Levine 2016) that results in a wider temporal activity range for the species, as individual swift foxes are actively balancing the tradeoff between the risk of mortality and the benefits of foraging, a pattern that is widely documented in traditional predator-prey systems (Verdolin 2006).When competitors are absent, the temporal overlap is high as all species are exploiting similar resources and using similar tactics that are optimized to a specific temporal window, but when dominant competitors are present, subordinates are displaced to less optimal foraging times, resulting in a wider temporal activity range for the species (Creel 2018, Monterroso et al. 2020, Ferreiro-Arias et al. 2021).The fossorial behavior of swift foxes may further facilitate such a reactive approach.Unlike coyotes and red foxes, which generally only associate with dens during the breeding season, swift foxes have several underground dens within their home range, where they tend to congregate (Sovada et al. 1998, Kitchen et al. 1999).The availability of a secure refuge may allow swift foxes to assess risk with few costs and, thus, enable swift foxes to coexist with coyote and red foxes despite significant spatial, temporal and ecological overlap (Kitchen et al. 1999, Cypher et al. 2001).
The seasonal change in temporal activity may further illustrate the dynamic nature of the inherent trade-offs between foraging and the risk of interspecific interactions.All three species exhibit substantially narrower activity breadth in the spring than in the fall and higher temporal overlap.Although it is possible that this pattern is simply a reflection of lower spring detection rates and thereby lower statistical power, it is worth noting that during the spring the energetic needs of parental care are increasing, but populations of prey are often near annual lows due to winter bottlenecks (Solonen 2006, Kaufman andKaufman 2018).The reduction in temporal breadth and increased temporal overlap could reflect a convergence by all three canid species to coincide foraging activity with the temporal availability and vulnerability of prey to increase foraging efficiency (Monterroso et al. 2014).For swift foxes, however, if a broad niche breadth is an adaptive response to the risk of interference competition, the reduction in niche breath suggests that the energetic needs of parental care are substantial enough to outweigh the costs of interference competition (Thompson and Gese 2012).Given that the cost of interference competition is mortality, such a trade-off seems unlikely, unless the risk of interference competition also changes seasonally.During the breeding season, home range sizes of coyotes and red foxes tend to decrease (Gosselink et al. 2003), creating a potential mosaic of safe space for swift foxes within the landscape and leading to a localized reduction in the risk of interference competition, independent of coyote abundance (Karki et al. 2007).Assuming swift fox can assess and respond to the risk of interference competition in the landscape (Kamler et al. 2003), proactive spatial partitioning of core areas may reduce the need for reactive temporal partitioning, a pattern that is consistent with other studies of swift foxes and coyotes spring activity patterns (Kitchen et al. 1999, Hertel et al. 2017).
In the fall, the pattern of partitioning is reversed, as the risk of interference competition increases (Karki et al. 2007) due to seasonal increases in population size and associated dispersal activity of young-of-the-year of all three canid species (Olson and Lindzey 2002, Olson et al. 2003, Finley et al. 2005, Martin et al. 2007).As the availability of safe space in the landscape decreases, swift foxes revert to dynamic temporal partitioning, a behavior that the diversity of abundant food resources may facilitate in the fall due to the prior summer season (Solonen 2006, Kaufman andKaufman 2018).Foxes and coyotes are opportunistic and dietary generalists that alter diet composition based on the availability and accessibility of prey (Kilgore 1969, Scott-Brown et al. 1987, Kitchen et al. 1999, Kamler et al. 2007).For example, swift foxes feed on small animals, especially rodents (e.g.prairie dogs, Cynomys ludovicianus) and rabbits (e.g.Sylvilagus spp.), but also small birds, such as meadowlarks (Sturnella spp.) and lark buntings (Calamospiza melanocorys) and bird eggs (Cutter 1958, Kilgore 1969, Uresk and Sharps 1986, Scott-Brown et al. 1987, Hines and Case 1991, Sovada et al. 2001, Kamler et al. 2007).Some of these swift foxes' prey exhibit a strong diurnal activity (e.g.prairie dogs) or activity concentrated at dawn and dusk (e.g.rabbits).The combined temporal activity pattern of multiple abundant prey species can provide continuous prey availability thought the day and night, potentially limiting the cost of foraging outside of optimal foraging times (Monterroso et al. 2014).Still, whether the foraging swift foxes can efficiently take advantage of such resources is unknown, and as such, so too are the fitness consequences.
Although swift foxes demonstrated the greatest degree of temporal activity breadth, all three species were largely cathemeral (i.e. the pattern of an organism's activity that occurs within both the light and dark portions of the daily cycle; Tattersall 1987, Eppley andDonati 2019).Like dominant carnivores, humans can also induce spatial and temporal shifts in wildlife (Laughrin 1977, Kitchen et al. 2000, Lesmeister et al. 2015).Our study occurred primarily on large ranches in a very rural landscape, where road densities and human activity were relatively low; therefore, foxes and coyotes may be less prone to avoid diurnal activity as the risk from human depredation may be low.There is evidence that coyotes' visual systems are best adapted to diurnal and crepuscular activity, and kit foxes (a closely related species to the swift fox) are better adapted to crepuscular light (Kavanau and Ramos 1975) and, therefore, these species are likely more effective in obtaining prey during these periods (Kitchen et al. 2000).With few constraints imposed by human activity, it may not be surprising that coyotes and swift foxes present more diurnal activity than expected.
Interactions among species that affect temporal activity patterns are difficult to understand because the activity pattern of a species is not only regulated by prey availability, competition and predation risk, but endogenous timekeeping mechanisms and other abiotic factors, such as environmental 1903220x, 2022, 6, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/wlb3.01027,Wiley Online Library on [10/11/2023].See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions)on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License light and darkness, weather events or ambient temperature (Kronfeld-Schor et al. 2013).Unique individual characteristics, such as age, sex, reproductive status and personality, may also shape activity patterns, consequently masking patterns at the population level (Hertel et al. 2017, Gaynor et al. 2018).Although our results are consistent with other studies of swift foxes (Kitchen et al. 1999) and theory (Creel 2018), we need to be cautious when interpreting activity curves generated from small sample sizes.Our survey protocol, which was by necessity designed at spatial and temporal scales to maximize swift foxes' detections, may have inadvertently obscured the scale at which temporal partitioning could be observed.We also assumed that our camera-trap detection rates represent the activity of our target species (Zimmermann et al. 2016), but the use of lures, camera site selection and even researcher activity may introduce error, further obscuring measures of activity patterns (Rowcliffe et al. 2008, Nouvellet et al. 2012).
Given the importance of activity patterns to ecological processes such as interspecific competition, efforts to understand species temporal patterns in different landscapes and communities using the appropriate scale framing are imperative.The relationship between organisms and their environment must be examined under the scheme of nature's heterogeneity on many spatial and temporal scales.Overall, our study just scratches the surface of the wide range of questions raised when trying to understand temporal activity patterns and the role of time as a niche axis.

Figure 1 .
Figure 1.Records of three canid species from camera traps.Time has been standardized to a day of equal length of day and night (sunrise at 06:00 h and sunset at 18:00 h).Dark grey bars represent records for the spring season (n = 980) and light grey bars for the fall season (n = 3391).The dashed lines represent mean value vectors.

Figure 2 .
Figure 2. Density estimates of daily activity for swift fox (Vulpes velox, yellow line, n = 2298), red fox (Vulpes vulpes, red line, n = 1306) and coyote (Canis latrans, blue line, n = 19 532) in western Nebraska, USA.The shaded yellow, red and blue areas represent 95% confidence intervals (CI); the shaded grey area represents the overlap of the three species density estimates.

Figure 3 .
Figure 3. Overall overlap plots of the density estimate of daily activity patterns for swift fox and coyote (a), swift fox and red fox (b) and red fox and coyote (c) in western Nebraska, USA.The yellow lines are density estimates for swift fox, red lines for red fox, whereas the blue lines are estimates for coyote.The dashed lines represent 95% confidence intervals (CI), and the activity coefficient of overlap (Δ) equals the grey shaded area below both curves.

Figure 4 .
Figure 4. Density estimates of daily activity patterns during two seasons for swift fox (a; spring n = 43, fall n = 399), red fox (b; spring n = 20, fall n = 206) and coyote (c; spring n = 917, fall n = 2786).The solid lines are density estimates for the spring, whereas the dashed lines are estimates for the fall.The yellow, red and blue shaded areas represent 95% confidence intervals (CI), and grey shaded areas represent the activity coefficient of overlap (Δ).

Figure 5 .
Figure 5. Overlap plots of the density estimates of daily activity patterns during two seasons for swift fox and coyote (a, b), swift fox and red fox (c, d) and red fox and coyote (e, f ) in western Nebraska, USA.The yellow lines are density estimates for swift fox, red lines for red fox, whereas the blue lines are estimates for coyote.The dashed lines represent 95% confidence intervals (CI), and the coefficient of overlap (Δ) equals the shaded area below both curves.

Table 1 .
Wald statistic on a chi-square distribution with one degree of freedom to test significant differences at the 5% level between overall activity patterns between spring and fall for each canid species.

Table 2 .
Estimate of activity pattern overlap (∆) between swift fox, red fox and coyote, sample size (n) and p-values.

Table 3 .
Wald statistic (difference, standard error, Wald test and p-values) on a chi-square distribution with one degree of freedom to test significant differences at the 5% level between overall activity patterns of three canid species.Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/wlb3.01027,Wiley Online Library on [10/11/2023].See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions)on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License considerable flexibility in activity patterns depending on the temporal patterns of prey, competitors and predators 1903220x, 2022, 6, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/wlb3.01027,Wiley Online Library on [10/11/2023].See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions)on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License