Author response:
The following is the authors’ response to the original reviews
Reviewer 1 (Public Review):
Summary:
In this paper, the authors aimed to test the ability of bumblebees to use bird-view and ground-view for homing in cluttered landscapes. Using modelling and behavioural experiments, the authors showed that bumblebees rely most on ground-views for homing.
Strengths:
The behavioural experiments are well-designed, and the statistical analyses are appropriate for the data presented.
Weaknesses:
Views of animals are from a rather small catchment area.
Missing a discussion on why image difference functions were sufficient to explain homing in wasps (Murray and Zeil 2017).
The artificial habitat is not really 'cluttered' since landmarks are quite uniform, making it difficult to infer ecological relevance.
Thank you for your thorough evaluation of our study. We aimed to investigate local homing behaviour on a small spatial scale, which is ecologically relevant given that the entrance of bumblebee nests is often inconspicuously hidden within the vegetation. This requires bees to locate their nest hole within a confined area. While many studies have focused on larger spatial scales using radar tracking (e.g. Capaldi et al. 2000; Osborne et al. 2013; Woodgate et al. 2016), there is limited understanding of the mechanisms behind local homing, especially in dense environments as we propose here.
We appreciate your suggestion to include the study by Murray and Zeil (2017) in our discussion. Their research explored the catchment areas of image difference functions on a larger spatial scale with a cubic volume of 5m x 5m x 5m. Aligned with their results, we found that image difference functions pointed towards the location of the objects surrounding the nest when the images were taken above the objects. However, within the clutter, i.e. the dense set of objects surrounding the nest, the model did not perform well in pinpointing the nest position.
See the new discussion at lines 192-197
We agree with your comment about the term "clutter". Therefore, we referred to our landmark arrangement as a "dense environment" instead. Uniformly distributed objects do indeed occur in nature, as seen in grasslands, flower meadows, or forests populated with similar plants.
See line 20 and we changed the wording throughout the manuscript and figures.
Reviewer 1 (Recommendations):
The manuscript is well written, nicely designed experiments and well illustrated. I have a few comments below.
It would be useful to discuss known data of learning flights in bumblebees, and the height or catchment area of their flights. This will allow the reader to compare your exp design to the natural learning flights.
In our study, we first focused on demonstrating the ability to solve a homing task in a dense environment. As we observed the bees returning within the dense environment and not from above it (contrary to the model predictions), we investigated whether they flew above it during their first flights. The bees did indeed fly above, demonstrating their ability to ascend and descend within the constellation of objects (see Supplementary Material Fig. 22).
In nature, the learning flight of bumblebees may cover several decametres, with the loops performed during these flights increasing with flight time (e.g. Osborne et al. 2013; Woodgate et al. 2016). A similar pattern can be observed on a smaller spatial scale (e.g. Philippides et al. 2013). Similar to the loops that extend over time, the bees gradually gain altitude (Lobecke et al., 2018). However, these observations come from studies where few conspicuous objects surround the nest entrance.
Although our study focussed on the performance in goal finding in cluttered environments, we now also address the issue of learning flights in the discussion, as learning flights are the scaffolding of visual learning. We have already conducted several learning flight experiments to fill the knowledge gap mentioned above. These will allow us in a forthcoming paper to compare learning flights in this environment with the existing literature (Sonntag et al., 2024).
We added a reference to this in the discussion (lines 218-219 and 269-272)
Include bumblebee in the title rather than 'bees'.
We adapted the title accordingly:
“Switching perspective: Comparing ground-level and bird’s-eye views for bumblebees navigating dense environments”
I found switching between bird-views and frog-views to explain bee-views slightly tricky to read. Why not use 'ground-views', which you already have in the title?
We agree and adapted the wording in the manuscript according to this suggestion.
I am not convinced there is evidence here to suggest the bees do not use view-based navigation, because of the following: In L66: unclear what were the views centred around, I assume it is the nest. Is 45cm above the ground the typical height gained by bumblebees during learning flight? The clutter seems to be used more as an obstacle that they are detouring to reach the goal, isn't it?
Based on many previous studies, view-based navigation can be assumed to be one of the plausible mechanisms bees use for homing (Cartwright & Collett, 1987; Doussot et al., 2020; Lehrer & Collett, 1994; Philippides et al., 2013; Zeil, 2022). In our tests, when the dense environment was shifted to a different position in the flight arena, almost no bees searched at the real location of the nest entrance but at the fictive new location within the dense environment, indicating that the bees assumed the nest to be located within the dense environment, and therefore that vision played a crucial role for homing. We thus never meant that the bees were not using view-based navigation. We clarified this point in the revised manuscript.
See lines 247-248, 250-259, added visual memory to schematic in Fig. 6
In our model simulations, the memorised snapshots were centred around the nest. However, we found that a multi-snapshot model could not explain the behaviour of the bees. This led us to suggest that bees likely employ acombination of multiple mechanisms for navigation.
We refined paragraph about possible alternative homing mechanisms. See lines 218-263
The height of learning flights has not been extensively investigated in previous studies, and typical heights are not well-documented in the literature. However, from our observations of the first outbound flights of bumblebees within the dense environment, we noted that they quickly increased their altitude and then flew above the objects. Since the objects had a height of 0.3 metres, we chose 0.45 metres as a height above the objects for our study.
Furthermore, the nest is positioned within the arrangement of objects, making it a target the bees must actively find rather than detour around.
I think a discussion to contrast your findings with Murray and Zeil 2017 will be useful. It was unclear to me whether the flight arena had UV availability, if it didn't, this could be a reason for the difference.
We referred to this study in the discussion of the revised paper (see our response to the public review). Lines 192-197
As in most lab studies on local homing, the bees did not have UV light available in the arena. Even without this, they were successful in finding their nest position during the tests. We clarified that in the revised manuscript. See line 334-336
Figure 2A, can you add a scale bar?
We added a scale bar to the figure showing the dimensions of the arena. See Fig. 2
The citation of figure orders is slightly off. We have Figure 5 after Figure 2, without citing Figures 3 and 4. Similarly for a few others.
We carefully checked the order of cited figures and adapted them.
Reviewer 2 (Public Review):
Summary:
In a 1.5m diameter, 0.8m high circular arena bumblebees were accustomed to exiting the entrance to their nest on the floor surrounded by an array of identical cylindrical landmarks and to forage in an adjacent compartment which they could reach through an exit tube in the arena wall at a height of 28cm. The movements of one group of bees were restricted to a height of 30cm, the height of the landmark array, while the other group was able to move up to heights of 80cm, thus being able to see the landmark array from above.
During one series of tests, the flights of bees returning from the foraging compartment were recorded as they tried to reach the nest entrance on the floor of the arena with the landmark array shifted to various positions away from the true nest entrance location. The results of these tests showed that the bees searched for the net entrance in the location that was defined by the landmark array.
In a second series of tests, access to the landmark array was prevented from the side, but not from the top, by a transparent screen surrounding the landmark array. These tests showed that the bees of both groups rarely entered the array from above, but kept trying to enter it from the side.
The authors express surprise at this result because modelling the navigational information supplied by panoramic snapshots in this arena had indicated that the most robust information about the location of the nest entrance within the landmark array was supplied by views of the array from above, leading to the following strong conclusions: line 51: "Snapshot models perform best with bird's eye views"; line 188: "Overall, our model analysis could show that snapshot models are not able to find home with views within a cluttered environment but only with views from above it."; line 231: "Our study underscores the limitations inherent in snapshot models, revealing their inability to provide precise positional estimates within densely cluttered environments, especially when compared to the navigational abilities of bees using frog's-eye views."
Strengths:
The experimental set-up allows for the recording of flight behaviour in bees, in great spatial and temporal detail. In principle, it also allows for the reconstruction of the visual information available to the bees throughout the arena.
The experimental set-up allows for the recording of flight behaviour in bees, in great spatial and temporal detail. In principle, it also allows for the reconstruction of the visual information available to the bees throughout the arena.
Weaknesses:
Modelling:
Modelling left out information potentially available to the bees from the arena wall and in particular from the top edge of the arena and cues such as cameras outside the arena. For instance, modelled IDF gradients within the landmark array degrade so rapidly in this environment, because distant visual features, which are available to bees, are lacking in the modelling. Modelling furthermore did not consider catchment volumes, but only horizontal slices through these volumes.
When we started modelling the bees’ homing based on image-matching, we included the arena wall. However, the model simulations pointed only coarsely towards the dense environment but not toward the nest position. We hypothesised that the arena wall and object location created ambiguity. Doussot et al. (2020) showed that such a model can yield two different homing locations when distant and local cues are independently moved. Therefore, we reduced the complexity of the environment by concentrating on the visual features, which were moved between training and testing (neither the camera nor the wall were moved between training and test). We acknowledge that this information should have been provided to substantiate our reasoning. As such, we included model results with the arena wall in the supplements of the revised paper. See lines 290-293, Figures S17-21
We agree that the catchment volumes would provide quantitatively more detailed information as catchment slices. Nevertheless, since our goal was to investigate if bees would use ground views or bird's eye views to home in a dense environment, catchment slices, which provide qualitatively similar information as catchment volumes, are sufficient to predict whether ground or bird's-eye views perform better in leading to the nest. Therefore, we did not include further computations of catchment volumes. (ll. 296-297)
Behavioural analysis:
The full potential of the set-up was not used to understand how the bees' navigation behaviour develops over time in this arena and what opportunities the bees have had to learn the location of the nest entrance during repeated learning flights and return flights.
Without a detailed analysis of the bees' behaviour during 'training', including learning flights and return flights, it is very hard to follow the authors' conclusions. The behaviour that is observed in the tests may be the result of the bees' extended experience shuttling between the nest and the entry to the foraging arena at 28cm height in the arena wall. For instance, it would have been important to see the return flights of bees following the learning flights shown in Figure 17. Basically, both groups of bees (constrained to fly below the height of landmarks (F) or throughout the height of the arena (B)) had ample opportunities to learn that the nest entrance lies on the floor of the landmark array. The only reason why B-bees may not have entered the array from above when access from the side was prevented, may simply be that bumblebees, because they bumble, find it hard to perform a hovering descent into the array.
A prerequisite for studying the learning flight in a given environment is showing that the bees manage to return to their home. Here, our primary goal was to demonstrate this within a dense environment. While we understand that a detailed analysis of the learning and return flights would be valuable, we feel this is outside the scope of this particular study.
Multi-snapshot models have been repeatedly shown to be sufficient to explain the homing behaviour in natural as well as artificial environments(Baddeley et al., 2012; Dittmar et al., 2010; Doussot et al., 2020; Möller, 2012; Wystrach et al., 2011, 2013; Zeil, 2012). A model can not only be used to replicate but also to predict a given outcome and shape the design of experiments. Here, we used the models to shape the experimental design, as it does not require the entire history of the bee's trajectory to be tested and provides interesting insight into homing in diverse environments.
Since we observed behavioural responses different from the one suggested by the models, it becomes interesting to look at the flight history. If we had found an alignment between the model and the behaviour, looking at thehistory would have become much less interesting. Thus our results raise an interest in looking at the entire flight history, which will require not only effort on the recording procedure, but as well conceptually. At the moment the underlying mechanisms of learning during outbound, inbound, exploration, or orientation flight remains evasive and therefore difficult to test a hypothesis. A detailed description of the flight during the entire bee history would enable us to speculate alternative models to the one tested in our study, but would remain limited in testing those.
While we acknowledge that the bees had ample opportunities to learn the location of the nest entrance, we believe that their behaviour of entering the dense environment at a very low altitude cannot be solely explained by extended experience. It is possible that the bees could have also learned to enter at the edge of the objects or above the objects before descending within the dense environment.
General:
The most serious weakness of the set-up is that it is spatially and visually constrained, in particular lacking a distant visual panorama, which under natural conditions is crucial for the range over which rotational image difference functions provide navigational guidance. In addition, the array of identical landmarks is not representative of natural clutter and, because it is visually repetitive, poses un-natural problems for view-based homing algorithms. This is the reason why the functions degrade so quickly from one position to the next (Figures 9-12), although it is not clear what these positions are (memory0-memory7).
In conclusion, I do not feel that I have learnt anything useful from this experiment; it does suggest, however, that to fully appreciate and understand the homing abilities of insects, there is no alternative but to investigate these abilities in the natural conditions in which they have evolved.
We respectfully disagree with the evaluation that our study does not provide new insights due to the controlled laboratory conditions. Both field and laboratory research are necessary and should complement each other. Dismissing the value of controlled lab experiments would overlook the contributions of previous lab-based research, which has significantly advanced our understanding of animal behaviour. It is only possible to precisely define the visual test environments under laboratory conditions and to identify the role of the components of the environment for the behaviour through targeted variation of them. These results yield precious information to then guide future field-based experiments for validation.
Our laboratory settings are a kind of abstraction of natural situations focusing on those aspects that are at the centre of the research question. Our approach here was based on the knowledge that bumblebees have to find their inconspicuous nest hole in nature, which is difficult to find in often highly dense environments, and ultimately on a spatial scale in the metre range. We first wanted to find out if bumblebees can find their nest hole under the particularly challenging condition that all objects surrounding the nest hole are the same. This was not yet clear. Uniformly distributed objects may, however, also occur in nature, as seen with visually inconspicuous nest entrances of bumblebees in grass meadows, flower meadows, or forests with similar plants. We agree that the term "clutter" is not well-defined in the literature and now refer to the environment as a "dense environment."
We changed the wording throughout the manuscript and figures.
Despite the lack of a distant visual panorama, or also UV light, wind, or other confounding factors inherent to field work conditions, the bees successfully located the nest position even when we shifted the dense environment within the flight arena. We used rotational-image difference functions based on snapshots taken around the nest position to predict the bees' behaviour, as this is one of the most widely accepted and computationally most parsimonious assessments of catchment areas in the context of local homing. This approach also proved effective in our more restricted conditions, where the bees still managed to pinpoint their home.
Reviewer 2 (Recommendations):
(1) Clarify what is meant by modelling panoramic images at 1cm intervals (only?) along the x-axis of the arena.
The panoramic images were taken along a grid with 0.5cm steps within the dense environment and 1cm steps in the rest of the arena. A previous study (Doussot et al., 2020) showed successful homing of multi-snapshot models in an environment of similar scale with a grid with 2cm steps. Therefore, we think that our scaling is sufficiently fine. We apologise for the missing information in the method section and added it to the revised manuscript. See lines 286-287
(2) In Figures 9-12 what are the memory0 to memory7 locations and reference image orientations? Explain clearly which image comparisons generated the rotIDFs shown.
Memory 0 to memory 7 are examples of the eight memorised snapshots, which are aligned in the nest direction and taken around the nest. In the rotIDFs shown, we took memory 0 as a reference image, and compared the 7 others by rotating them against memory 0. We clarified that in the revised manuscript.
See revised figure caption in Fig. S9 – 16
(3) Figure 9 seems to compare 'bird's-eye', not 'frog's-eye' views.
We apologise for that mistake and carefully double-checked the figure caption.
See revised figure caption Fig. S9
(4) Why do you need to invoke a PI vector (Figure 6) to explain your results?
Since the bees were able to home in the dense environment without entering the object arrangement from above but from the side, image matching alone could not explain the bees’ behaviour. Therefore, we suggest, as an hypothesis for future studies, a combination of mechanisms such as a home vector. Other alternatives, perhaps without requiring a PI vector, may explain the bees’ behaviour, and we will welcome any future contributions from the scientific community.
References
Baddeley, B., Graham, P., Husbands, P., & Philippides, A. (2012). A Model of Ant Route Navigation Driven by Scene Familiarity. PLoS Computational Biology,8(1), e1002336. https://doi.org/10.1371/journal.pcbi.1002336
Capaldi, E. A., Smith, A. D., Osborne, J. L., Farris, S. M., Reynolds, D. R., Edwards, A. S., Martin, A., Robinson, G. E., Poppy, G. M., & Riley, J. R. (2000).
Ontogeny of orientation flight in the honeybee revealed by harmonic radar. Nature, 403. https://doi.org/10.1038/35000564
Cartwright, B. A., & Collett, T. S. (1987). Landmark maps for honeybees. Biological Cybernetics, 57(1), 85–93. https://doi.org/10.1007/BF00318718
Dittmar, L., Stürzl, W., Baird, E., Boeddeker, N., & Egelhaaf, M. (2010). Goal seeking in honeybees: Matching of optic flow snapshots? Journal of Experimental Biology, 213(17), 2913–2923. https://doi.org/10.1242/jeb.043737
Doussot, C., Bertrand, O. J. N., & Egelhaaf, M. (2020). Visually guided homing of bumblebees in ambiguous situations: A behavioural and modelling study. PLoS Computational Biology, 16(10). https://doi.org/10.1371/journal.pcbi.1008272
Lehrer, M., & Collett, T. S. (1994). Approaching and departing bees learn different cues to the distance of a landmark. Journal of Comparative Physiology A, 175(2), 171–177. https://doi.org/10.1007/BF00215113
Lobecke, A., Kern, R., & Egelhaaf, M. (2018). Taking a goal-centred dynamic snapshot as a possibility for local homing in initially naïve bumblebees. Journal of Experimental Biology, 221(2), jeb168674. https://doi.org/10.1242/jeb.168674
Möller, R. (2012). A model of ant navigation based on visual prediction. Journal of Theoretical Biology, 305, 118–130. https://doi.org/10.1016/j.jtbi.2012.04.022
Murray, T., & Zeil, J. (2017). Quantifying navigational information: The catchment volumes of panoramic snapshots in outdoor scenes. PLOS ONE, 12(10), e0187226. https://doi.org/10.1371/journal.pone.0187226
Osborne, J. L., Smith, A., Clark, S. J., Reynolds, D. R., Barron, M. C., Lim, K. S., & Reynolds, A. M. (2013). The ontogeny of bumblebee flight trajectories: From Naïve explorers to experienced foragers. PLoS ONE, 8(11). https://doi.org/10.1371/journal.pone.0078681
Philippides, A., de Ibarra, N. H., Riabinina, O., & Collett, T. S. (2013). Bumblebee calligraphy: The design and control of flight motifs in the learning and return flights of Bombus terrestris. Journal of Experimental Biology, 216(6), 1093–1104. https://doi.org/10.1242/jeb.081455
Sonntag, A., Lihoreau, M., Bertrand, O. J. N., & Egelhaaf, M. (2024). Bumblebees increase their learning flight altitude in dense environments. bioRxiv, 2024.10.14.618154. https://doi.org/10.1101/2024.10.14.618154
Woodgate, J. L., Makinson, J. C., Lim, K. S., Reynolds, A. M., & Chittka, L. (2016). Life-long radar tracking of bumblebees. PLoS ONE, 11(8). https://doi.org/10.1371/journal.pone.0160333
Wystrach, A., Mangan, M., Philippides, A., & Graham, P. (2013). Snapshots in ants? New interpretations of paradigmatic experiments. Journal of Experimental Biology, 216(10), 1766–1770. https://doi.org/10.1242/jeb.082941
Wystrach, A., Schwarz, S., Schultheiss, P., Beugnon, G., & Cheng, K. (2011). Views, landmarks, and routes: How do desert ants negotiate an obstacle course? Journal of Comparative Physiology A: Neuroethology, Sensory, Neural, and Behavioral Physiology, 197(2), 167–179. https://doi.org/10.1007/s00359-010-0597-2
Zeil, J. (2012). Visual homing: An insect perspective. Current Opinion in Neurobiology, 22(2), 285–293. https://doi.org/10.1016/j.conb.2011.12.008
Zeil, J. (2022). Visual navigation: Properties, acquisition and use of views. Journal of Comparative Physiology A. https://doi.org/10.1007/s00359-022-01599-2
