Study Area
Hampstead Heath is a 275-hectare urban green space, which is listed as a site of Metropolitan Importance for Nature Conservation and situated in a densely populated part of Central London (City of London, 2019). Hampstead Heath comprises a mosaic of habitats including woodland, grassland, scrub and open water, which are managed to maintain a diverse and biodiverse landscape while ensuring access by visitors (City of London, 2010; City of London, 2019). Wildlife surveys show woodlands, including an ancient woodland designated as a Site of Special Scientific Interest (Natural England 1990), to be particularly rich in breeding bird and mammal species (City of London, unpublished).
Data Collection
Between 18th April and 5th July 2018, camera traps were uniformly distributed over Hampstead Heath across 150 sites using a 150 m2 cell-size grid overlay (Monterroso et al, 2013). Each camera trap operated on average for 15 days (range: 9 - 23 days) and recorded continuously 24 hours a day. Survey lengths varied due to limited storage on memory cards, batteries running out of charge and camera failures. Camera traps were attached 20-50 cm above the ground to suitable features nearest to the ideal coordinates, for example fence posts or trees. Traps were angled slightly downwards to increase the probability of recording wildlife. Traps minimised disturbance by passively detecting wildlife using heat signatures (Welbourne et al, 2016) and were not baited. 104 camera traps across four models were used: 64 Browning Strike Force HD-Pro BTC-131 5HDPs, 20 Reconyx HyperFire HC-500s, 10 Reconyx Hyperfire HC-600s and 10 Reconyx Hyperfire PC-800s. The Browning model took one image per trigger with a delay of approximately 0.6 seconds between photos, while the Reconyx models took ten images per trigger with a delay of one second (specifications for these cameras can be found at www.trailcampro.com).
Camera Site Selection
For the purpose of this study, a subset of 26 camera traps were selected from woodland sites across Hampstead Heath (Online Resource-Table 1). Camera traps were selected in woodland designated as “at least 84% broadleaved woodland”, with the exception of one camera in a wooded copse in “semi-neutral improved grassland” (City of London, 2009; Online Resource-Table 2). Broadleaved woodland sites were dominated by English oak (Quercur robur) and less frequently by common beech (Fagus sylvatica) and sessile oak (Quercus petraea), with an understory of bramble (Rubus fruticosus), holly (Ilex aquifolium) and other small trees, and a ground cover of common ivy (Hedera helix) or bracken (Pteridium aquilinum) (See Online Resource-Table 2 for full habitat descriptions by site). Within this habitat, all camera traps were positioned to give a clear view of open leaf litter to five metres with little ground cover under a closed canopy. This improved the detection probability for all wildlife species, especially hedgehogs which can go undetected in tall vegetation (Sollmann et al, 2013). Importantly, the distribution of cameras across the Heath meant some were close to, and others distant from, foot paths. Some were in the few fenced off woodland areas on the Heath, which people and dogs rarely visited. This created a broad range in frequency of human and dog visits recorded between traps.
Photo Review
All vertebrate species within camera images were manually identified using ExifPro v2.1 (Kowalski, 2013) and classed as ‘detections’ and/or ‘contacts’. When an animal enters a camera’s field of view, the camera triggers and takes a series of successive images of the animal. Each successive image of that animal was classed as a ‘detection’, while the first image, indicating the arrival of a new individual, was classed as a ‘contact’. Hence a contact was often the first image in a series of detections for an individual of a particular species. While the sum of all detections from a specific trap gives a measure of the total time spent by that species at that site, the sum of contacts estimates the total number of individuals of that species visiting the site. To estimate contacts, we assumed that an interval of two or more minutes between successive detections of a species represents the arrival of a new individual (Kay and Parson, 2014). For wildlife species, humans and dogs, we recorded total contacts as an indicator of visits to a particular site. Dogs on and off leads were both counted as contacts.
Wildlife Selection
Across the 26 camera traps, images of 18 distinct species of wildlife were captured, with some species being particularly common (Table 1). Diurnal species included a range of ground-foraging birds and grey squirrels (Sciurus carolinensis), while nocturnal species included mostly red foxes (Vulpes vulpes) and European hedgehogs (Erinaceus europaeus). For our analysis of wildlife activity, we put species in groups (All Birds and Non-diurnal Mammals) and also investigated individual species where we had enough detections.
Table 1 Summary of wildlife species detected in camera images across the 26 sites, with the number of absolute and relative detections (out of total 13,617 wildlife detections), and the number of sites at which the species was detected in camera images.
Common Species Name
|
Latin Name
|
Total Number of Detections
|
Proportion of Wildlife Detections
|
Total Sites
|
Common Blackbird
|
Turdus merula
|
1,646
|
12%
|
19
|
Magpie
|
Pica pica
|
999
|
7%
|
21
|
Song Thrush
|
Turdus philomelos
|
478
|
4%
|
13
|
Wood Pigeon
|
Columba palumbus
|
956
|
7%
|
12
|
Other Birdsa
|
-
|
501
|
4%
|
18
|
European Hedgehog
|
Erinaceus europaeus
|
482
|
4%
|
18
|
Grey Squirrel
|
Sciurus carolinensis
|
6,402
|
47%
|
26
|
Red Fox
|
Vulpes vulpes
|
1,769
|
13%
|
24
|
Other Mammalsb
|
-
|
386
|
3%
|
10
|
Table 1 footnotes: a ’Other birds’ consists of blue tits (Cyanistes caeruleus), carrion crows (Corvus corone), dunnocks (Prunella modularis), Eurasian jays (Garrulus glandarius), great tits (Parus major), European robins (Erithacus rubecula) and wrens (Troglodytes troglodytes), which individually have very few tags and often were present at only a few sites.
b ’Other mammals’ consists of brown rats (Rattus norvegicus), wood mice (Apodemus spp.), European rabbits (Oryctolagus cuniculus) and muntjacs (Muntiacus reevesi), which individually have very few tags and were present at only a few sites.
More frequently observed species were also better represented over a range of sites with different levels of humans and dog visits. Based on total detections, we selected four bird species for individual analysis: common blackbirds (Turdus merula), magpies (Pica pica), song thrushes (Turdus philomelos), and wood pigeons (Columba palumbus). We also created for analysis the category “All Birds” in order to incorporate these species and the individually rare but collectively significant diversity of “Other Birds” (Table 1). Three species of mammals were analysed individually: red foxes, grey squirrels and European hedgehogs. Because mammal species were, broadly speaking, either diurnal, crepuscular or nocturnal, and likely respond to differently to daytime visits by humans and dogs (Gaynor et al, 2018), we created for analysis a group of “Non-diurnal Mammals” comprising red foxes, European hedgehogs, wood mice (Apodemus sylvaticus), European rabbits (Oryctolagus cuniculus) and muntjac (Muntiacus reevesi). Grey squirrels were the only diurnal mammal species analysed. Camera trap records of brown rats showed them to be predominantly diurnal (Online Resource-Figure 1b), but their numbers were so few that they were excluded from the analysis.
Statistics
Establishing sites of low and high human and dog visits
Human and dog contacts, separately recorded at each site, were summed and divided by the time of operation of the camera at each site to generate Relative Abundance Indices (RAIs) in contacts per day (Carbone et al, 2001). We justify the use of these “Human + Dog RAIs” on the basis that dogs were not seen without humans, and the effects of both on wildlife could not be easily separated. Human and dog visits were highly correlated (simple linear regression with logged RAIs: b=0.67, p-value<0.01, R2=0.70; Online Resource-Table 3 and Figure 2).
Human + Dog RAIs per site ranged from zero for cameras in remote, fenced areas to 127 contacts day-1 for cameras near busy paths. Examining all sites, we identified a natural breakpoint between four and eight human and dogs contacts per day (Reilly et al, 2017; Online Resource-Figure 3). Sites with less than five humans and dogs contacts per day were defined as low visit sites (n = 17, median = 2 contacts day-1), while sites with five or more human and dogs contacts per day were defined as high visit sites (n = 9, median = 13 contacts day-1).
Spatial Displacement
To determine potential spatial displacement, RAIs were also calculated for wildlife species at each site. Wildlife RAIs were put into low and high visit site groups and the two groups compared using Wilcoxon tests, due to the general non-homogeneity of variances, lack of normality and unequal group sizes.
Temporal Displacement
To examine temporal displacement of wildlife by human and dog visits, we calculated at each site the degree of temporal overlap between Human + Dog detections and the detections for different wildlife species, and then compared these results between our low and high visit sites. We assumed that at low visit sites, where human and dog disturbance is less, the observed activity schedules are more likely to reflect the natural times of activity for a species. Hence, reduced temporal overlap at high visit sites would mean that species are being displaced from their natural activity schedules (Reilly et al, 2017).
To calculate temporal displacement, detections were used, rather than contacts, as the method requires total species activity and does not need to distinguish individuals (Rowcliffe et al, 2014). Camera models differed in photographic rate, meaning that they generated different numbers of detections in response to a species crossing in front of the camera. By filtering the data so that only detections recorded at least one minute apart were included in the analysis, these camera differences were removed (note: measuring spatial displacement did not require correcting for camera differences as it utilised contacts, which are at least two minutes apart).
Timestamps were then converted to sunTime to standardise how sunrise and sunset varied between 18th April and 11th June across the 26 sites: times at sunrise and sunset were standardised to π/2 and 3π/2 respectively. This corrected for changes in timings of behaviour due to the seasonal photoperiod change (Nouvellet et al, 2012; Meredith and Ridout, 2020).
Temporal overlap estimates were generated using the R package ‘Overlap’ (Ridout and Linkie, 2009; Meredith and Ridout, 2020). ‘Overlap’ converts sunTime, in radians, into non-parametric kernel density estimates for each species, i.e. activity schedules (Ridout and Linkie, 2009; Meredith and Ridout, 2020). The activity schedule across 24 hours of each species is then superimposed with that generated for combined humans and dogs. The area underneath where the two activity schedules overlap is the temporal overlap estimate, which has a value between 0 and 1 that represents no and complete temporal overlap respectively (Ridout and Linkie, 2009; Meredith and Ridout, 2020).
To be selected for individual analysis, a species needed to have at least ten detections to produce robust estimates of temporal activity (Díaz‐Ruiz et al, 2016). If either species (i.e. a wildlife species or humans + dogs) in the temporal overlap analysis had less than 75 detections, density estimates were generated using the Dhat1 estimator with a smoothing coefficient of 0.8 (Meredith and Ridout 2020). This estimator compensates for the reduced number of detections by over smoothing the activity schedule to increase its overall accuracy (Ridout and Linkie, 2009). Alternatively, the Dhat4 estimator was used which has a smoothing coefficient of 1. To estimate the accuracy of the temporal overlap estimates, we generated 95% confidence intervals by recalculating the temporal overlap estimates 1,000 times using bootstrapping (Meredith and Ridout, 2020).
Finally, temporal overlap estimates between each wildlife species and humans and dogs were compared between low and high visit sites using Wald Chi-Squared tests. The formula for which is as follows:
where TH and TL are the temporal overlap estimates from high and low visit sites, respectively, and seH and seL are the standard errors of these estimates from high and low visit sites, respectively. Using a degree of freedom of one, temporal overlap estimates from low and high visit sites were statistically significant if the chi-squared value was greater than 3.841 (p-value < 0.05).
All plotting and analyses were conducted in R v.3.6.2 (R Core Team, 2019).