Panorama similarity and navigational knowledge in the nocturnal bull ant Myrmecia midas

Solitary foraging ants acquire navigational information from both the surrounding terrestrial landmarks and their path integrator (Wehner, 2008; Cheng, 2012; Collett et al., 2013; Freas and Schultheiss, 2018). Navigation via path integration involves coupling a pedometer with the use of multiple celestial cues including the position of the sun and the polarised light pattern present in the sky (Wehner, 2008; Wehner and Müller, 2006; Reid et al., 2011; Zeil et al., 2014; Freas et al., 2017a). Ant navigators will rely on these celestial cues when their habitat lacks terrestrial landmarks or when foragers are less experienced with the terrestrial makeup of a location (Wehner and Srinivasan, 2003; Bühlmann et al., 2011; Freas and Cheng, 2017). When terrestrial information is available, foragers also acquire and retain terrestrial cue information from the surrounding panorama to navigate between locations (Collett et al., 2006; Collett, 2010; Narendra, 2007; Freas et al., 2017b,c). Before an individual begins foraging, it first acquires views of the terrestrial panorama around the nest through a series of learning walks (Zeil, 2012; Jayatilaka et al., 2013, 2018; Fleischmann et al., 2016, 2018a,b; Freas et al., 2019). During these learning walks, pre-foraging ants of both Cataglyphis noda and Myrmecia croslandi are known to occasionally look back in the direction of the nest, likely learning the surrounding panorama, and these walks can extend to over 2 m from the nest entrance (Fleischmann et al., 2018b; Jayatilaka et al., 2018).

Panorama learning also occurs along the individual's foraging route, with foragers developing robust memories of non-nest sites (Nicholson et al., 1999; Kohler and Wehner, 2005; Graham and Cheng, 2009; Zeil, 2012; Schultheiss et al., 2016; Freas et al., 2017b; Freas and Cheng, 2018; Freas and Spetch, 2019). As a forager travels away from the nest, it will also occasionally turn and look back in the nest direction (Nicholson et al., 1999; Zeil, 2012; Zeil et al., 2014). Memories of the nest panorama and the panorama along the foraging route allow the forager to return to the nest by comparing their current view while navigating with memories of the panorama at the nest and along the foraging route (Collett et al., 2006; Wehner et al., 2006; Philippides et al., 2011; Baddeley et al., 2012; Zeil, 2012; Zeil et al., 2014; Kodzhabashev and Mangan, 2015; Ardin et al., 2016; Murray and Zeil, 2017).

The range or catchment area at which stored panoramas can provide navigational information to a foraging ant is dependent, at least in part, on the clutter of the environment (Stürzl and Zeil, 2007; Murray and Zeil, 2017). When there are many nearby landmarks, such as in a densely wooded area, the catchment area will be smaller compared with a more open environment comprising more distant landmarks (Zeil et al., 2003; Stürzl and Zeil, 2007; Murray and Zeil, 2017) because panorama similarity is dependent on the visibility of distant landmarks, which are more constant over longer distances. In more cluttered environments where nearby landmarks block distant cues, panoramas of the nest catchment area or other known sites will be smaller compared with open habitats where distant landmarks are visible and unobstructed (Murray and Zeil, 2017). As the degree of habitat clutter mediates panorama similarity, the navigational success of animal navigators in cluttered environments should be dependent on both the surrounding landmark density and the navigational experience of the individual forager away from the nest. However, ant research into the catchment area around the nest has focused on species that live in cluttered but open environments where distant landmarks remain visible to foragers, despite the clutter. Research on the Australian desert ant Melophorus bagoti and the diurnal bull ant M. croslandi has shown that foragers return to the nest after displacement to previously unvisited sites of up to 10 m away, suggesting high similarity of the panorama within a 10 m radius of the nest site (Wystrach et al., 2012; Narendra et al., 2013). Yet in another bull ant species, Myrmecia midas, foragers that do not move away from the nest area on the ground are unable to orient to the nest after small local displacements (5 m; Freas et al., 2017c) despite actively navigating along the vertical foraging route on the nest tree (Freas et al., 2018).

In this study, we explore whether navigational performance in M. midas foragers in highly cluttered environments is dependent on panorama similarity between the release site and previously visited locations. We characterised the navigational knowledge of three forager groups according to how far from the nest they forage, the first being ‘nest tree foragers’, which do not travel on the ground away from the nest but instead forage in the nest tree. The second group, ‘close tree foragers’, comprised individuals that travelled relatively short distances on the ground to a nearby tree to forage (2.5 m at nest 1 and 3 m at nest 2). The final group, ‘far tree foragers’, comprised individuals that travelled long distances along the ground to reach their foraging tree (14 m at nest 1 and 8 m at nest 2). Foragers from each of these groups were displaced off-route at different distances from the nest entrance, and their initial orientation and homeward paths were collected. We conducted image analysis of the nest panorama and the panoramas en route to explore if scene similarity between release site panoramas and known panoramas was linked to the observed differences in navigational performance in these forager groups.

The current study was conducted on Myrmecia midas Clark 1951 from November 2017 to March 2018 on the Macquarie University campus in Sydney, Australia (33°46′18″ S, 151°06′30″ E). The two nests studied were ∼275 m apart, with foragers travelling nightly to nearby trees within a 15 m (typically ≤5 m) radius of the nest. M. midas at this field site nest in wooded areas with stands of eucalyptus trees and largely barren understories. Nests are located close to the base of a tree and a portion of the foraging force (∼30%) travels up this tree (nest tree foragers) while the remaining foragers travel to surrounding trees. An individual's preferred foraging tree has been observed to remain stable over a period of at least a few weeks, with foragers travelling to the same foraging tree each night. However, foragers at the nest tree have been occasionally observed to switch to other foraging trees. As forager group classifications are based on the forager's observed natural behavioural range, we cannot with certainty know each forager's complete experience around the local area. Collection and testing occurred during twilight, so we utilised headlamps with red filter covers, which did not appear to affect the ants. Studying and collecting ants in Australia requires no ethical approval at the local, state or federal level and all collection and testing procedures were non-invasive, with no observed effects on individuals or the nest as a whole. Additionally, after foragers completed testing by either returning to the nest area or leaving the testing area, they were returned to the nest site where all foragers freely entered.

At both nests, 360 deg reference panoramas were collected (Theta 360 deg camera, Ricoh Company) near the nest entrance (30 cm) and at the midpoints of each foraging route. At nest 2, separate reference panoramas at the nest were taken from each displacement direction (one nest panorama taken from the tree and close forager displacement direction and one nest panorama taken from the far forager displacement direction). Each nest panorama was centred on the nest entrance from the release site direction, while foraging route midpoint panoramas were centred on the nest direction. Additionally, nest-centred 360 deg panoramic images were collected at each of the 11 release sites. Rotational image mismatches were calculated between each release site panorama and the three reference panoramas at nest 1 and the two reference panoramas at nest 2. Each panoramic image was downsized to 360 pixels in width and 80 pixels in height (approximately 60 pixels above and 20 pixels below the horizon). To reduce the effects of cloudy portions of the sky, these images were converted to greyscale by retaining only the blue colour channel. Rotational image difference functions (rotIDFs) were calculated by using the sum of the absolute difference in pixel intensity between the release site panoramas and the reference images, for all possible rotations (in one-degree steps) using custom-written scripts in MATLAB (Zeil et al., 2003, 2014; Stürzl and Zeil, 2007).

From the nest entrance, the close foraging tree and far foraging tree were directionally separated by 60 deg. Three release sites were marked 30 deg off-route, directionally equidistant from both trees, at 2.5 m, 6.5 m and 10.5 m from the nest entrance. From these sites to the nest entrance a 5 m×12 m grid of 1 m×1 m squares was erected using string and metal stakes with the grid ending 50 cm from the nest entrance (Fig. 1A). Foragers typically need to search for the exact location of the nest entrance once they are in the nest area and the 50 cm gap between the grid and the nest entrance was in place as we ignored this search behaviour in the current study.

To determine the navigational knowledge of foragers travelling different distances to their foraging tree, at nest 1, we separated foragers into three testing groups: (1) nest tree foragers, (2) close tree foragers and (3) far tree foragers. Using a plastic vial, foragers were collected during the evening twilight as they climbed onto their foraging tree. Nest tree foragers were collected on the tree located 20 cm from the nest entrance, close tree foragers at a tree 2.5 m from the nest, and far tree foragers at a tree 14 m from the nest. Foragers were marked with a small amount of enamel paint (Tamiya) with the colour corresponding to the foraging tree. In order to mark foragers, individuals were cold anaesthetised at −18°C in a commercial freezer for 3 min. This procedure has been previously shown to not adversely affect foragers' ability to navigate using terrestrial cues (Freas, 2015). Marked foragers were fed with a small amount of honey, held overnight in a darkened container and released back at the base of their foraging tree the following morning, ensuring each forager had at least one experience of both the outbound and inbound route between the nest and foraging tree. During the evening twilight, any marked foragers observed leaving the nest were followed for 1 m to ensure they were travelling towards the same foraging tree and were collected using a plastic vial. Nest tree foragers were collected as they reached approximately 1 m in height on the tree face. These foragers were marked as tested using a distinct paint colour, fed a small amount of honey, held overnight and then tested the following night, during dusk twilight. Collected foragers were tested at one of the release sites using a 40-cm-diameter reference circle painted on a wooden board. At the centre of the reference circle was a 2-cm-diameter hole which was connected to a 2-cm-deep plastic container beneath the surface of the board. Foragers were deposited into the container and allowed to climb up and on to the board's surface. Each forager was allowed to travel off the board and onto the ground. The forager's initial heading as it crossed the reference circle at 20 cm was marked using a pin and recorded using string and a digital compass. Foragers were then allowed to travel off the board where their homeward paths were recorded using a pencil and graph paper. Foragers that returned to the nest were allowed to enter, while foragers that travelled off the 5 m×12 m grid were allowed 1 min to return to the grid area or were collected and returned to the nest entrance where they all freely entered.

At nest 2, foragers were again separated into the three testing groups. Identically to nest 1, foragers were collected during the evening twilight as they climbed onto their foraging tree using a plastic vial. Nest tree foragers were collected on the tree located 14 cm from the nest entrance, close tree foragers at the tree 3 m from the nest, and far tree foragers at the tree 8 m from the nest. Nest 2 was located at the base of two trees separated by ∼15 cm. Given the proximity of the trees to each other and the nest and because ants from this nest foraged in both, foragers ascending either tree were identified as nest tree foragers.

At nest 2, the foraging tree configuration was unsuitable for a single group of release sites between the close and far foraging tree directions. At this nest, two separate groups of release sites were marked 45 deg counter-clockwise off-route from the straight-line route from nest to foraging tree at 2.5 m, 4.5 m, 6.5 m and 10.5 m from the nest entrance (Fig. 1B). From these sites to the nest entrance a 5 m×12 m grid was erected ending 50 cm from the nest entrance (Fig. 1B). The foraging tree 3 m from the nest was located on the nest side of the nest tree while the 8 m foraging tree was located on the opposite side of the nest tree (Fig. 1B). Since nest tree foragers may show reduced success when attempting to navigate from opposite the nest direction, we chose to displace this group at the release sites extending out from the nest entrance side of the tree.

At nest 2, foragers were initially collected similarly to nest 1, except that at nest 2, foragers were individually marked. Marked foragers were then fed, held overnight and released at their foraging tree during the subsequent pre-dawn twilight. When a marked forager was observed leaving the nest and travelling in the direction of its foraging tree, it was collected after 1 m. These foragers were fed, held overnight and tested during the morning twilight at the 2.5 m site where their initial headings and homeward paths were recorded. At nest 2, each marked forager was tested at all four release sites beginning with the 2.5 m site, with tests increasing to the 4.5 m, 6.5 m and 10.5 m sites, with one test per collection. This non-random testing design was implemented to ensure foragers had no previous experience of closer release sites from previous testing displacements. After each test, foragers were allowed to return to the nest and were collected on their next observed foraging trip. Foragers that were observed travelling to a foraging tree that did not correspond with their marking were excluded from the study. (In actuality, one marked nest tree forager was observed travelling to and climbing onto the 8 m foraging tree and was collected and marked to prevent testing.)

To compare foragers' navigational performance at each site to the depth of the valley of mismatch in the panorama, the rotIDF minimum depth was calculated by subtracting the minimum pixel-difference value from the 95th percentile value (Fig. 2; Fig. 3, dashed lines) for each site for both comparisons to the nest site and en route site reference images. For nest tree foragers, the rotIDF minimum depth from each release site to the nest site was used. As both close tree and far tree foragers would have memories of both the nest site panorama and the panorama en route to their foraging tree, the larger rotIDF minimum depth of the two was considered the best match and used. Pearson's correlation coefficients with Bonferroni corrections (α set at P=0.0125) were used to test the association between these rotIDF minimum depths and four metrics, two metrics of navigational performance and two metrics of navigational error: (1) the percentage of foragers successfully returning to the nest, (2) the percentage of forager initial orientations within ±23 deg of the rotIDF minimum direction, (3) the circular variance of forager initial orientations and (4) the degree of error between the mean vector of initial orientations and the direction predicted by the rotIDF minimum directions.

Initial orientations were analysed with circular statistics (Batschelet, 1981; Zar, 1998) using Oriana Version 4 (Kovach Computing Services). We analysed forager orientations to determine if they met the conditions of a uniform distribution using Rayleigh Tests (P>0.05), and we analysed if non-uniform data were significantly grouped around the nest direction using V-tests (α set at P=0.05). Additionally, to further analyse if foragers' initial orientations were grouped towards the nest, we examined if the nest direction (0 deg) was within the 95% confidence interval (CI) of observed headings.

At both nest 1 and nest 2, when comparing the 2.5 m release site panoramas to their respective reference panoramas, rotIDFs show a distinct best match direction near the nest direction at 1 deg and 354 deg, respectively (Table 1; Fig. 2A,B; Fig. 3A–D). Additionally, comparing the rotIDF minimum depth of the 2.5 m site with both the nest panorama and panoramas en route indicated the nest site panorama comparison had the greatest rotIDF minimum depth (Table 1).

Behavioural data at both nests showed foragers in all three groups successfully oriented to the nest direction with initial orientations that were non-uniform and grouped in the nest direction (Table 2; Fig. 4A–C,J–L; see Table S1 for raw data). Additionally, 8 of 10 (80%) nest tree foragers, 10 of 10 (100%) close tree foragers, and 10 of 10 (100%) far tree foragers successfully found the nest entrance at nest 1 (Fig. 5A–C) and 13 of 15 (86.7%) nest tree foragers, 15 of 15 (100%) close tree foragers, and 15 of 16 (93.8%) far tree foragers successfully found the nest entrance at nest 2 (Fig. 6A–C).

On the nest tree forager and close tree forager side of nest 2, when comparing the 4.5 m release site panorama to the nest panorama, the rotIDFs show a less distinct valley near the nest direction with the minimum at 352 deg (Table 1, Fig. 3A,C; pink) and a decrease in the nest panorama's rotIDF minimum depth (Table 1). In comparison, rotIDFs between the release point and the en route reference panorama showed a clear valley of mismatch at 27 deg (Fig. 3A,C; blue) and larger rotIDF minimum depths (Table 1). On the far tree forager side of nest 2, both the nest panorama (pink) and far route panorama (blue) contained clear valleys of best match near the nest direction at 8 deg and 60 deg, respectively (Fig. 3B,D), but the en route panorama contained the greatest rotIDF minimum depth (Table 1).

Behavioural data mirrored these changes in the panorama, as nest tree foragers were no longer oriented to the nest and exhibited initial headings in a uniform distribution (Table 2; Fig. 4M), although a majority of foragers, 9 of 15 (60%), returned home successfully (Fig. 6D). Both close tree foragers and far tree foragers remained well oriented to the nest direction (Table 2; Fig. 4N,O), and 14 of 15 (93.3%) and 13 of 15 (86.7%) of these individuals successfully returned to the nest, respectively.

The rotIDF best match with both the nest and close en route reference panoramas continued to degrade as release site distance increased. Analysis at the 6.5 m and 10.5 m sites no longer showed a distinct valley in the nest direction at both nest 1 (Fig. 2A,B) and the nest/close tree foragers side of nest 2 (Fig. 3A,C) for either reference panorama (Table 1). RotIDF minimum depth at both release sites was greatest when compared with the nest panorama at nest 1 and the en route panorama at nest 2 (Table 1).

In contrast, the nest 1 reference panorama along the 14 m foraging tree route panorama (yellow) provided the best match near the nest direction with both the 6.5 m and 10.5 m release site panoramas, at 40 deg and 38 deg, respectively (Table 1; Fig. 2A,B). Yet at nest 2, both the nest panorama (pink) and en route panorama (blue) contained clear valleys of best match when compared with release site panoramas (Table 1; Fig. 3D). The nest site panorama contained the greatest rotIDF minimum depth during comparisons with the 6.5 m site, while the en route panorama contained the greatest rotIDF minimum depth during comparisons with the 10.5 m site (Table 1).

Foragers' navigational success corresponded with these panorama similarities. Nest tree foragers released at either 6.5 m or 10.5 m exhibited uniform initial headings and were not grouped towards the nest direction (Table 2; Fig. 4D,G,P,S) and only 3 of 10 (30%) nest 1 foragers (Fig. 5D) and 7 of 15 (46.7%) nest 2 foragers (Fig. 6G) successfully returned home from the 6.5 m site (one forager at nest 1 climbed up the close foraging tree after searching around its base and was collected and returned to the nest). At 10.5 m, only 3 of 10 (30) nest 1 foragers (Fig. 5G) and 1 of 15 (6.7%) nest 2 foragers (Fig. 6J) successfully returned home.

Close tree forager navigational performance at both nests also declined at the 6.5 m and 10.5 m sites, exhibiting uniform initial orientations no longer oriented to the nest site (Table 2; Fig. 4E,H,Q,T), although at 6.5 m, a majority of foragers still successfully returned to the nest: 7 of 11 (63.6%) nest 1 foragers (Fig. 5E) and 12 of 15 (80%) nest 2 foragers (Fig. 6H). At the 10.5 m release site, only 4 of 10 (40%) nest 1 foragers (Fig. 5H) and 2 of 15 (13.3%) nest 2 foragers (Fig. 6K) successfully returned to the nest.

Far tree foragers remained well oriented regardless of the displacement distance. Forager initial orientations were non-uniform and grouped around the nest direction (Table 2; nest 1, Fig. 4F,I; nest 2, Fig. 4R,U). At nest 1, all foragers (10) successfully returned to the nest at the 6.5 m release site (Fig. 5F) and 9 out of 10 (90%) returned from the 10.5 m site (Fig. 5I). At nest 2, 12 of 15 (80%) returned from the 6.5 m site (Fig. 6I) and 10 of 15 (66.7%) returned from the 10.5 m site (Fig. 6L).

RotIDF minimum depth magnitude significantly correlated with multiple navigational metrics (Fig. 7). As rotIDF minimum depth increased, navigational success in both initial headings and successfully returning to the nest improved (Fig. 7A) and both circular variance in initial headings (Fig. 7B) and error between heading mean vector and the rotIDF minimum direction decreased (Table 1; Fig. 7C).

In the current study, we show that the navigational performance of M. midas foragers corresponds with the degree of similarity between the release site panorama and the panoramas of previously visited sites. Nest tree foragers were shown to orient successfully only 2.5 m from the nest tree, where a large nest rotIDF minimum depth exists. Beyond this distance, both the nest rotIDF minimum depth and navigational success decreases. Close tree foragers show high navigational success up to the 4.5 m release site where nest and en route rotIDF minimum depths at the 4.5 m site remain high. However, both rotIDF minimum depth and navigational performance decrease at longer distances (6.5 m and 10.5 m). Far Tree Foragers exhibited the best navigational performance at the 6.5 m and 10.5 m release sites, which corresponded to large en route rotIDF minimum depths, even at these distances. Additionally, the best-match rotIDF minimum depth correlated positively with navigational success and negatively with heading variance and heading error. Taken together, our results suggest that the cluttered environmental makeup in which individuals forage leads to decreases in panorama similarity between the nest panorama and release site panoramas over short distances. Panorama similarity between known panoramas and the release site panoramas appears to be linked with navigational success based on an individual's previous knowledge of locations away from the nest.

It is important to note that in this study the panorama images taken at the midpoint on the foraging route are not meant to represent the only stored views of close and far tree foragers away from the nest. Foragers likely acquire multiple panoramic memories along their foraging route and previous work in these foragers has shown that foraging routes between the nest and foraging tree are not in perfectly straight lines (Freas et al., 2017a). Instead, at the midway point between the nest and foraging tree, the variation of the foraging route between individuals can be 2 m wide (at nest 1), before narrowing closer to the foraging tree. Foragers in these conditions may have experience much closer to the release sites than Fig. 1 suggests. As such, the en route panorama and the rotIDF minimum depth comparisons between the release sites and this panorama should be seen as general proxy for panorama memories beyond the nest rather than as a single nest-directed panorama memory.

As a side note to the image analysis, the rotIDF minimum depth in the far tree forager displacement on the opposite side of the nest tree (Fig. 3B,D) at nest 2 suggests that in this instance the nest panorama may still contain similarities to the panorama at the 4.5 m and 6.5 m site, with the rotIDF minimum depths for the nest site and en route being quite similar (Table 1) at these distances. This was likely due to a clearing beyond the nest tree in this direction, leading to a less cluttered panorama. It is possible that nest tree foragers when displaced to these sites may have exhibited better navigational performances. A future study might attempt to explore individual M. midas nest habitats that are more open and may better facilitate nest tree forager navigation from more distant release sites.

Additionally, while the current findings generally support the hypothesis that nest tree foragers do not have experience beyond the nest panorama, this does not appear to be the case for all individuals. A small sub-set of nest tree and close tree foragers do orient correctly home and exhibit relatively straight nestward paths similar to their far foraging nestmates. This suggests some level of variability in forager knowledge and that some of these individuals may have travelled further from the nest entrance than would be expected from their observed foraging behaviour during testing. A long-term foraging ecology study would be useful in untangling when foragers begin to forage beyond the nest area.

An alternative explanation for the differences in navigational performance observed in the current study posits that close and far tree foragers have access to both a 1 m accumulated vector and long-term vector memories of the foraging route, which could help them orient and navigate back to the nest. This explanation is unconvincing to us, as M. midas foragers have been shown previously not to orient to vectors under 5 m and only weakly orient to vector cues with 14-m-long accumulated vectors (Freas et al., 2017c). Additionally, foragers were displaced off their foraging route (30 deg at nest 1 and 45 deg at nest 2), and if foragers were employing vector cues to navigate there should be some bias towards the foraging tree-to-nest direction in either the initial orientations or the nestward paths. Our results show no evidence of this bias at either nest, suggesting that any effect of these vector cues would be minimal. Furthermore, this explanation would not explain the differences observed between the close and far tree foragers, both of which would have access to these cues.

A final interesting aspect of this study concerns the potential learning of the off-route panoramas during testing at nest 2. As foragers have experience of homeward trips from previous tests during the 4.5 m, 6.5 m and 10.5 m displacements, it is possible that successful foragers would have access to these memories during future tests, through reinforcement learning (Freas et al., 2019). Yet, this is not supported by differences in orientation success between nests 1 and 2, with both nest tree foragers and close tree foragers being unable to orient successfully at the 6.5 m displacement at both nests. If foragers learned and retained route memories of previous tests, nest 2 foragers would be expected to exhibit successful orientation from the 6.5 m and 10.5 m sites. Furthermore, this testing setup does not appear to explain the differences between our testing groups, as learning previous inbound routes should aid all foragers during testing at the further release sites. Panorama memories acquired while returning to the nest have been shown in other ant species to be weak, and it is possible that foragers may have trouble retaining long term memories of panoramas experienced only during the inbound route even after multiple experiences (Freas and Cheng, 2017, 2018; Freas and Spetch, 2019).

In conclusion, our findings show that M. midas foragers in highly cluttered environments exhibit different levels of navigational success based on the amount of panorama similarity between release locations and the known sites. Foragers with no experience away from the nest showed decreased navigational success corresponding with decreases in similarity between the nest panorama and release site panoramas. Foragers with knowledge of foraging routes away from the nest performed better and their navigational performance corresponded with panorama similarity levels between the foraging group's en route panorama and release site panoramas.

We thank Macquarie University for access to the field site. We would like to thank Marcia Spetch for her comments on a draft of the manuscript. We also thank Antoine Wystrach for his help with the rotIDF analysis.