1. Introduction
Southern hairy-nosed wombats (
Lasiorhinus latifrons) inhabit a fragmented distribution across the semi-arid regions of south-central South Australia and the south-eastern corner of Western Australia (
Figure 1). While recent studies have been able to map the species-wide distribution of southern hairy-nosed wombats using very high-resolution satellite imagery [
1], determining how many wombats there might be is more problematic [
2,
3,
4]. Wombat abundance at the broad-scale is estimated by using a proxy measure of counting the number of active burrows (burrows which are currently being used by wombats), which is then multiplied by an index of the number of wombats/active burrow. However, there are several problems with this approach.
Indices of the number of wombats/active burrow were first calculated in the 1980s, based on mark-recapture studies using cage traps installed at burrow entrances, which were conducted near Fowlers Bay on the west coast of South Australia [
5] and near Blanchetown in the Murraylands [
6]. The number of individual wombats caught in the traps were counted, and the result was divided by the number of active burrows being surveyed. While hairy-nosed wombats generally have a preferred warren in which they reside for the majority of the time, they also use up to ten different warrens throughout their home range [
7]. As a result, indices of the number of wombats/active burrow based on how many individual wombats there may be in a warren could potentially over-estimate the index by counting not only the wombats which regularly reside in the warren, but also wombats who were visiting from nearby warrens. The use of invasive approaches such as cage traps and mark-recapture surveys also have the potential to skew the results, as wombats will adopt measures such as remaining in their burrows for up to 10 days to avoid the traps [
8].
We hypothesise that, rather than attempting to identify the number of wombats in an area and then relating that to the number of active burrows (i.e., the object of the study being the number of wombats), a more accurate assessment of the number of wombats/active burrow can be obtained by determining the proportion of active burrows which are being simultaneously occupied by wombats (i.e., the object of the study is the burrow occupancy rate). To achieve this, a reliable method of determining when wombats enter and exit each burrow, and how many wombats are using the burrow at the same time, is required; one which does so non-invasively to minimise any changes in the animals’ behaviour.
Motion-activated infra-red cameras (camera traps/wildlife trail cameras) offer a relatively cheap and logistically simple means of monitoring wildlife behaviour [
9]. The use of motion-activated cameras for wildlife research in Australia has increased significantly over the past decade, with an exponential increase in the number of published papers since 2010 describing the use of camera traps [
10]. While much of the increase in use of motion-activated cameras can be attributed to improvements in the capability and reductions in the cost of the cameras, improvements in both the capacity and cost of data storage has also been important [
11]. Wildlife trail cameras can now capture and store tens of thousands of high-resolution day or night-time images, meaning that they can be left in place longer and can collect more data than earlier camera models. The growth in data storage capacity has also facilitated the greater use of video to analyse wildlife behaviour. This has allowed for a greater range and a more accurate interpretation of animal behaviour than would otherwise be possible using still imagery alone [
12]. However, while still imagery has been used on a number of occasions for wombat research [
4], there has been no published research on the use of video to monitor wombat behaviour in the wild.
The aims of this study were to compare the effectiveness of video and still imagery for determining how often the active burrows of southern hairy nosed wombats are utilised, and to use video and still imagery to calculate an index of the number of wombats/active burrow based on burrow occupancy rates.
4. Discussion
This study informs two important areas of wombat population research; the reliability of cameras as tools for monitoring wombat activity, and the calculation of an index of the number of wombats/active burrow for use in population abundance estimates. Our results provide evidence for the efficacy of video as a tool for determining warren occupancy rates, albeit with some limitations. They also provide lessons in regard to understanding how the cameras work, and ensuring that they are properly sited and the correct settings are used.
Comparisons between the video and still imagery collected during phase one of this study indicated that there was a clear difference in our ability to positively identify wombat burrow occupancy from video versus the still imagery. The identification of burrow occupancy from video was relatively simple, whilst the approach used in previous studies of identifying individual wombats from still imagery can be difficult and time consuming, with identification relying on characteristics (nose hair patterns, scars, ear notches, etc.) that are not always visible in the imagery. Further, because wombats do not just reside in one warren, but rather use up to ten different warrens throughout their home range [
7], accounting for wombats which might be ‘just visiting’ a warren, as opposed to being a resident of that warren, is problematic. We have cases where we have counted over 30 different wombats using 14 burrows over a period of one month; which equates > 2 wombats/active burrow. Consequently, counting wombats would clearly require more detailed statistical analysis than just assessing burrow occupancy in order to produce accurate abundance estimates.
The failure of the long-range cameras to contribute much useful information during phase one of the study (25% of burrow occupancies identified) highlights some of the limitations with using cameras for this type of research [
15]. There were numerous occasions when data was captured on the close-up and medium-range cameras, but nothing was recorded on the long-range cameras. This suggests that the distance (~10 m) may have been beyond the cameras’ range, despite the manufacturers claiming a detection range of up to 20 m. We suggest that this is probably only applicable to larger animals, as we did not observe wombats at anything approaching that distance.
The Passive Infra-Red (PIR) sensors used on these cameras detect the surface temperature of objects in the field of view, and trigger the camera when a change in temperature is detected in part or all of the sensor [
16]. This occurs when an object with a different surface temperature to its surrounds changes its aspect in relation to the sensor (i.e., moves across the field of view). However, the cameras sometimes trigger when there is no obvious target within their field-of-view. We noted that this phenomenon was particularly prevalent during the daytime, with 86% of all false positive detections that we recorded occurring in the afternoon, and less than 2% occurring at night. Although we did not analyse the false positive rate for phase two, given that we set the cameras to revert to standby mode between 10:30 and 14:30 each day, we noticed that there were few false positive video recordings on any of these cameras, and of the few that did occur, almost all were between 14:30 and 18:00.
Whilst previous studies have suggested that these false positives may be triggered by the wind moving vegetation within the field-of-view of the camera [
17], our results suggest that other factors may have a greater effect. For example, we noticed that while false positives did occur frequently on windy days, they also occurred at a similar rate on still days when neither the camera nor branches were in motion. The increasing number of false positives which occurred as the day progressed and the lower rate of false positives recorded during the night suggests that it may be solar radiation causing the uneven heating of the ground in front of the camera, or heating of the camera body itself, which causes the cameras to trigger. We recommend that further research and pre-survey trials be conducted before undertaking future wildlife studies of this nature, to verify the causes and rate of false positives for the particular camera model being used.
The high rate of false positives, coupled with the data requirements of high-definition (720 P) video, meant that we used up the data storage capacity (32 Gb) of some of the video cameras before the end of the planned survey period during phase one of the study. As a result, we reduced the resolution of the videos to the minimum size possible on the cameras (320 × 480 P) to reduce the data storage requirements for phase two of the study. While this largely solved the problem of data capacity, the high volume of wombat traffic at one site (Dakalanta site #2) still resulted in the data capacity of the cameras at that site being used up after 20 days. This highlights the need to carefully evaluate the potential volume of data which may be captured by a camera and the data storage capacity available, and to adjust the camera settings, survey periods and logistics of any study to ensure that data is not missed as a result of inadequate data storage capacity.
In regard to the siting of the cameras around warrens, in the case of a wombat emerging from its burrow there would be only a limited aspect change in relation to the camera oriented directly into the burrow (i.e., the close-up camera). The surface temperature of the wombat’s body is also likely to be similar to the ambient temperature of the burrow, especially when the wombat is covered in dirt and the weather is warm. This suggests that fossorial species such as wombats may represent a poor target for PIR detection. There was evidence for this in our observations of burrow entry/exit. Although we generally observed the entire burrow entry process from the video (the wombat approaching, descending into the burrow crater and entering the burrow), in the case of burrow exits the camera usually did not trigger until the wombat had completely emerged from the burrow and was climbing out of the crater (i.e., we rarely observed the wombat inside the mouth of the burrow). We also noted cases where imagery of a wombat exiting the burrow would be captured on the medium-range camera, which was oriented side-on to the burrow entrance, but no imagery was captured on the close-up camera (
Figure 6). This phenomenon of cameras failing to detect animals which were within their field of view has been previously described, with up to 68% of verified animal activity being missed in some circumstances [
17].
Whether the failure of some of our cameras to detect wombat activity which was occurring within their field of view—and which was captured on other cameras with a different orientation to the target—is an inherent limitation of wildlife trail cameras, or whether it is a limitation of these type of cheaper, generic cameras which are entering the market is difficult to tell from our data. In either case, our findings provide strong support for the importance of understanding the capabilities and limitations of motion-activated infra-red cameras in general, and of the model of camera being used, prior to their use in studies of this type. It also underscores the importance of the correct placement and alignment of the camera in relation to the target and to minimise the potential for false triggers to occur [
15]. Unfortunately, it may not be possible to overcome all the potential limitations with a single camera no matter its quality or how well it is sited, and hence multiple cameras located in different positions may be necessary in some situations. If this is the case, it raises questions about the findings from previous studies where only one camera was used to detect animal presence/activity, and whether some data might have been missed as a result. We recommend further research in this area.
In regard to our calculation of the number of wombats/active burrow, our figure of 0.43 for the Murraylands is virtually identical to previous figures calculated for the region of 0.43 [
6], 0.4389 [
2,
13] and 0.40 [
18]. The figure of 0.42 for the Eyre Peninsula is the first index which has been calculated for that area. Both figures are lower than 0.60 calculated for the Far West Coast [
5] and 0.50 which was used for the only species-wide abundance estimates undertaken to date [
19].
Whilst other studies suggest that a range of indices are required for different conditions such as soil type, especially for small-scale population estimates in areas with a large number of calcrete warrens [
4,
20], the remarkable similarity between the figures we calculated for the Murraylands and Eyre Peninsula suggests that, at the broad scale at least, an index of ~0.43 might be a robust figure to calculate wombat abundance. Nonetheless, as this issue is fundamental to our understanding of wombat abundance at both the local and broad scales, we would recommend on-going research to validate these findings, especially in regions other than those previously surveyed and/or under a range of environmental conditions such as drought [
13,
21].
Our results also highlight the difficulties with accurately assessing whether a burrow is active, an issue which has been previously identified in multiple studies [
6,
13,
22]. Two survey sites, site #1 at Moorunde and site #1 at Dakalanta were both assessed as active, based on recent diggings and fresh wombat scats around the burrow entrances. However, whilst we observed wombat activity in the vicinity of the burrow entrance at both sites throughout several nights during the survey period, as well as activity by other species including kangaroos and feral goats, there was no evidence on the imagery of a wombat entering or exiting either burrows at any time during the survey. It appears that activity by both visiting wombats and non-target species around the entrance to the burrows disturbed the soil, and together with the deposition of scats, resulted in the misidentification of the burrows’ status. We also noted the phenomenon at other survey sites. For example, although the burrow occupation rate for Dakalanta site #5 was only 0.08 (burrow occupied for 2 out of 25 survey days), the signs—digging, footprints, fresh scats—around the burrow entrance suggested a high rate of activity. This was supported by the video evidence, which showed wombat activity in the vicinity of the burrow, including digging around the burrow entrance, on all but two nights (23/25 nights) of the survey period.
Whether these inaccuracies in the subjective assessment of a burrow’s active status constitutes a problem depends upon the context. Certainly, at the local scale, if research effort is spent on a burrow which is not active, then that effort may be wasted. Conversely, as the calculated indices of the number of wombats/active burrow are all based on subjective assessments which also includes the misidentification of the status of some burrows [
2,
5,
6,
13], we do not consider this to be a significant problem for broad area population studies. Nonetheless, consideration should be given to using small scale studies as models to develop indices prior to any broad-scale population survey, and to ensure that subjective assessments by multiple researchers are standardised to reduce potential errors in the abundance calculations.
We also noted days where no wombat activity was captured on the cameras—and hence, we did not observe a wombat entering or exiting the burrow—but we assessed that the burrow was occupied. This can be explained by the observation of a wombat entering the burrow but not emerging until several days later. We noted three instances of a wombat remaining in its burrow for more than one day, with the longest period being three and a half days (wombat entered at 05:07 on 29 November and did not emerge until 21:29 on 2 December). A second wombat also did not emerge from a different burrow for two days during the same period. This was most likely related to the weather, with our cameras recording maximum temperatures in excess of 40 °C on both 29 and 30 November 2017. A herd of feral goats was also observed grazing on the surface of the warren on the evening of the 30 November (
Figure 7). This avoidance behaviour by wombats of high temperatures and outside disturbances has been previously described by multiple sources [
13,
23,
24,
25,
26]. It also highlights the merits of non-invasive approaches which reduce the potential to cause behavioural changes, and of surveys which investigate burrow occupancy rather than wombat activity, which is likely to vary seasonally and in response to environmental conditions.
In regard to burrow sharing by more than one wombat, while this is not thought to be common, it does occur, especially in areas dominated by layers of sheet calcrete limestone [
27] where access to the subterranean shelter may be limited to breaks or fractures sites in the limestone layer. Previous studies have also suggested that these warrens can have more complex underground structures; with large open areas between the calcrete layers and several chambers which are accessed from a single burrow entrance [
28]. We noted four sites (out of 18 total survey sites) where a single burrow entrance was used simultaneously by more than one wombat; Moorunde sites #4 and #5, and Dakalanta sites #2 and #3. For three out of four of these sites, burrow sharing occurred on only a few days out of the survey period. At Dakalanta site #2 burrow sharing occurred across two-thirds of the survey period (12 out of 18 days), with three wombats being observed to share the burrow for one of those days. As a consequence, we would recommend that further research be undertaken to ascertain burrow sharing behaviour and kinship relationships among adult wombats [
29,
30].
With studies of this nature there is always the possibility that the cameras did not capture all the wombat activity, Although the number of cameras that we used reduces the likelihood of this occurring, we cannot discount the possibility that we may have missed one or more burrow occupancies. As a consequence, our indices of 0.43 for the Murraylands and 0.42 for the Eyre Peninsula should be considered to be minimum figures, and further research should be undertaken to verify these findings.