Camera deployment
Cameras were set at three sites in the Elkhorn Plain within Carrizo National Monument: 1) 35.197505°,−119.71552° (named site 2); 2) 35.163363°,−119.674° (named site 5); and 3) 35.11432°,−119.6209° (named site 7) (Fig. 1). Sites were at least 2 km apart and at least 1 km2. Sites located on public land held by the U.S. Bureau of Land Management were selected based on known presence of the target species (blunt-nosed leopard lizards), and presence of shrubs in the area to allow shrub-open paired microsites for camera deployment. The daily movement of blunt-nosed leopard lizards usually ranges between 65 and 110 m but can reach up to 300 m [36]. Therefore, the 2 km spacing between sites was selected to provide a sufficient buffer for extreme dispersal events of this and other species, to ensure that small mammals and lizard sightings at one site on a given day were independent of sightings at other sites.
A total of 14 Primos camera traps (Primos Hunting, Flora, MS, USA) and 6 Reconyx camera traps (Reconyx, Holmen, WI, USA) were deployed from April 1st through July 5th, 2015. Both models have a camera of at least 3 megapixels (3.1 MP for the Reconyx, 7 MP for the Primos), with at least 2 sensitivity settings (5 for the Reconyx, 2 for the Primos), and have replaceable batteries. Both models use Secure Digital (SD) cards. Both cameras also contain a passive infrared (PIR) motion sensor for detecting heat-in-motion during the day and night [37, 38]. Cameras were deployed in a paired design with 2 cameras focused on microsites located near a shrub; one camera facing the north side of a shrub between 1 and 2 m from the shrub, the other facing away from the shrub focused on an open area between 5 and 10 m away from the north side of the surveyed shrub. This area was at least 5 m away from any other shrub. Cameras were attached to 20 cm pegs anchored firmly in the ground, with the camera placed between 10 and 15 cm off the ground. Vegetation was left intact and disturbance was minimized. From April until mid-May, cameras were set at each shrub-open pair from sunrise to sunset, and then moved to a new shrub-open pair the next day, for a total of 133 shrub-open pairs sampled during this time. From mid-May to July cameras were deployed in a pattern of one, 4-day deployment followed by two, 1-day deployments. During this time 49 shrub-open pairs were sampled for 4 days, and 61 shrub-open pairs were sampled for 1 day. All deployments were at new shrub-open pairs so as to more intensively sample microhabitats at peak animal activity, while still sampling as many shrub-open pairs as possible. In some instances cameras were deployed for slightly shorter or slightly longer periods of time, depending on when they were able be checked. Exact dates and lengths of each deployment for each camera are listed in the metadata.
Imagery collected
Sites were surveyed on consecutive days, but the order in which they were surveyed was random. At each site, shrubs were also selected at random to survey as many shrubs as possible over the course of the season. Individual shrubs were not surveyed more than once during the study, to avoid repeated measures and to ensure that desert sites were sampled as widely was possible. At each set of paired shrub-open sites, shrub size (length, width, and height), and distance to nearest 3 shrubs was recorded. Annual plant abundance, annual plant species composition, and annual plant density were recorded using a 0.25 m2 quadrat at the location of each camera. Camera settings (picture mode, number of pictures per trigger, and sensitivity level) were also recorded. Cameras were set to record a three-picture burst per trigger followed by a 10 s delay at the highest quality picture setting. The default (highest) sensitivity level setting was used. Camera images numbering 425,369 in Joint Photographic Experts Group (JPEG) format were examined to determine the presence of animals over the season. False triggering of the cameras by grass or other vegetation moving in the wind was common, leading to greater battery power consumption and rapid filling of the memory storage cards. However, this did not lessen the effectiveness of the camera, as animal movement still triggered the camera sensors. Battery life and data storage could last more than 1 week, so no survey time was lost to false triggers. We reviewed 100,000 of the images for content before uploading them to GigaDB to ensure image quality, presence of animals, date and temperature stamp, and data integrity.
Metadata
Animals were detected in 0.4 % of the reviewed images, including San Joaquin antelope squirrels (Ammospermophilus nelson), jackrabbits (Lepus californicus), coyotes (Canis latrans), blunt-nosed leopard lizards (Gambelia sila), whiptail lizards (Cnemidophorus tigris), side-blotched lizards (Uta stansburiana), loggerhead shrikes (Lanius ludovicianus), and grasshopper and butterfly species (Fig. 2). Images from cameras at both shrub and open microsites were clear and readable, and the camera resolution was sufficient under challenging conditions (such as low light) to discern and identify animals during both day and night (Fig. 3). The mean file size for each JPEG image was 750 kb. Filenames were encoded based on the location of the camera, the camera number, and the date the pictures were taken. Metadata provided for each file include the start and end date of the collection, site id, shrub number for that day, whether the camera was at a shrub or open microsite, the direction the camera was facing, the camera number and type, camera settings, and the geographic location. A selection of images containing animals are provided along with the raw imagery data to provide an example of the species detected. Even though the cameras were placed in a desert ecosystem with high ambient temperatures, the cameras did not exceed their maximum operating temperature, and there were no camera failures. It is nonetheless possible that at very high desert temperatures, detection rates might be influenced by background infrared interference during daylight hours [39].
Potential uses
Camera traps can provide data for a range of basic ecology and wildlife ecology research. A common use of camera traps is to assess the presence or absence of animal species [1, 2]. Cameras are able to survey continuously, which is thus useful in the detection of rare or elusive species [40]. Imaged animals can be identified to the species level, allowing an inventory of species to be created [2]. As with other types of trapping, the method requires an adequate number of days to examine species presence. This can vary by species, but is usually around 30 days [1, 2. 40].
Camera traps are also used to estimate relative species abundance according to the number of times each species triggers the camera-the more triggers by a particular species, the greater is its relative abundance in the area being surveyed [2]. There are, however, limitations to this estimate because an individual animal can trigger multiple images during a short time period or over several days [2, 40]. If specific individuals can be identified from the images, then mark-recapture analysis can be used to obtain a more accurate estimate of abundance [2]. In this type of model, the frequency of marked or known individuals and the frequency of unmarked or new individuals are used to calculate abundance [2, 40]. Though useful, this method is limited because there is potential for misidentification. Not all individuals have identifying marks, or the marks may be hidden, or require subjective classification [40]. Species density may be found using abundance [2, 40]. However, to do this requires calculation of the effective trapping area, or area covered by the cameras [40].
Even if individuals of a species cannot easily be identified, abundance and density may still be calculated using a random encounter model [41]. This method uses the species’ trapping rate, the speed of the captured animal, and the location of the animal relative to the camera when captured to estimate abundance and density [40, 41]. This model assumes that animals move randomly and independently so its effectiveness varies by species [40].
Cameras can be deployed in different habitat types within an area to determine a species’ habitat use [2] and develop a habitat occupancy model. These models can be used to estimate occupancy of habitat by the species of interest, as well as its detectability within that habitat [42]. A selection of sites is sampled to estimate these values over a larger site or area. These models require repeated observation of the study sites [42]. Many cameras also have a time-lapse function, which allows them to be used to monitor a site over time [1]. Taking pictures at set time intervals can be useful for detecting changes at the site over time, whether those changes are animal activity, vegetation conditions, or weather events [1]. The field of camera trapping continues to evolve, and potential for the use and reuse of camera imagery datasets will only increase [2].
With this dataset, our goal was to determine whether the blunt-nosed leopard lizard was present in the study area of Elkhorn Plain. In addition, we wanted to determine if camera traps were an effective tool for detecting this particular species. The paired shrub-open design also allows habitat and microsite use to be examined with the dataset. Cameras successfully detected animal species in the open and were also effective under shrub canopies. With subsequent analyses, these data can be used to estimate microhabitat occupancy, and-to a lesser extent-spatial co-occurrence patterns. The animal/insect capture rate of 0.4 % suggests that extensive temporal and spatial sampling is required, particularly if the target animal species is relatively rare [5]. However, extended surveys of changes in populations and activity over time are possible because the camera traps can be deployed for a week or more with little to no maintenance [2, 37, 38]. Though this survey was only carried out between spring and summer, seasonal differences in animal abundances and densities may be assessed from the dataset. Additional cameras can also be easily deployed to extend the scale of the survey and sample a wider area [37, 38]. The inclusion of a date and time stamp on all of the pictures taken mean that time periods of specific activity for this animal species can also be described [43–45]. Limited, direct species behavior can also be described depending on how long animals spend within the frame of the camera-when animals remain in front of the camera for some time, cameras are triggered to take photographs frequently enough that animal activity and length of activity can be determined. Because this dataset provides information about the presences or absence of species at our study site, it may be useful as a starting point for researchers conducting more in depth surveys on the Elkhorn Plain. In addition, this dataset could be combined with other imagery datasets from the area to provide a more comprehensive picture of animal distribution and activity.
Image data can be useful in describing vegetation characteristics and short-term micro-environmental disturbances-we also propose that changes in vegetation over time may be described. Weather events such as rainfall and cloud patterns are recorded by these cameras. Temperature is encoded into every image, and these data can be extracted and used to evaluate ambient temperature differences between microenvironments and sites. Hence, image data from camera traps can be relevant to general ecology and vegetation science as well as conservation and wildlife biology.
Despite all their benefits, imagery datasets can be difficult to manage because of their sheer size. Manually processing hundreds of thousands of images is a time consuming process [8]. An automated approach, such as an automated algorithm, would greatly reduce the time and effort required for image processing. Imagery datasets such as this one would be useful as a training dataset for these algorithms as they are continually developed and improved.
Availability of supporting data
The imagery dataset is deposited in the GigaDB repository [46], including the image, camera, and study design metadata. Images (in JPEG format) are organized into folders based on the specific deployment site and date. Basic information about the habitat at the camera microsite, the type of camera trap used at that microsite and its settings, and the location of the survey site are also provided.
Ethics approval
Disturbance to plants and animals was kept to a minimum throughout the study. No plants or animals were collected during the course of the study. Authorization for this project was granted by Carrizo Plain National Monument, U.S. Bureau of Land Management, 3801 Pegasus Drive, Bakersfield, CA 93308–6837, USA.