Using Machine Learning to Identify Associations between the Environment, Occurrence, and Outcomes of Songbird Displacements at Supplemental Feeders

Round 1

Reviewer 1 Report

This is a minute paper, in terms of data collection and number of areas/temporal periods considered. In other words, quite arrogantly the authors attempt to exploit an AI and machine learning approach to detail both intra- and inter- specific interactions in birds. In particular intraspecific networks cannot be revealed by such a methodology.

 

The way authors use to measure aggressive interactions, mainly based on bioacustical indexes is definetly weak.  Also the strategy exploiting experimental feeders may induce major biases eg due to their actual positioning and relative distance: it is rather difficult to use bird feeders as standardized methodology to collect field data. Of course any attempt to implement new methodologies can be considered useful. However, their  feeble statements, the not enlisted yet numerous biases, etc… should be clearly reported in a “honest” (in my personal opinion) final conclusion.

 

Author Response

Please see the attachment

Author Response File: Author Response.docx

Reviewer 2 Report

General comments:

This is an interesting study exploring the combined use of automated logging and predictor identification in field conditions. Although the authors identify throughout the text that this is a proof-of-concept work and that broad hypotheses were purposefully adopted, there are two ways in which this feature can be improved. First, the introduction does not make this clear from the beginning and the reader is left to wonder why the authors provide superficial mentions of various, seemingly unrelated factors of aggressive feeding interactions and displacement events. I suggest the authors frame the work from the onset in terms of utilising modern automated approaches to inform our understanding of predictor complexity in the field and to better characterise events such as displacements, highlighting also the benefits of their approach (i.e. the PASSER random-forest coupling). Second, even though this is a proof of concept study, it does deliver some interesting findings in terms of what drives displacement events and roles. Thus, the manuscript would benefit from a better interpretation of findings in the context of existing evidence on displacement behaviour factors, but also a better justification or a re-examination of predictor selection for the two models. For instance, contest theory literature can inform the authors on the social-environment effects but could also inform predictor selection, e.g. resource value factors or aggressive-ability factors such as the size of displaced and displacer (see Arnott & Elwood, 2008 & 2009). Overall, I consider this approach very useful in studying the complexity of aggressive interactions in the field, and this could potentially expand to examining a number of conceptual frameworks on ethology, ecology and evolution, which should be better communicated by the text after addressing current minor concerns.

Specific comments:

L73-81: The authors should provide some definition of what is a behavioural syndrome. Further, in the context of what is described here, it seems unsurprising that a boldness syndrome comprising “proactive approach” as per the authors’ words is correlated with the tendency to initiate displacement interactions. So the authors need to clarify what is the significance of this here but also the link, existing or proposed, between the morphological factors, behavioural-correlates and ecological factors.

 

L104-106: How does the hypothesised effect of other’ presence here differ from the effect of the number of individuals from previous studies reported by the authors in L78-81?

 

L93-109: Given the intent to address the “broader physical and social environment” effects (L94), can the authors explain why the focus on certain factors. For instance, why environment and time-of-day factors, and not other surrounding features such as cover, location, accessibility feeder size etc. Or why the presence of con- and hetero- specifics and not interaction dynamics, familiarity, age or dominance disparities, sex etc.

 

L120-13: Here it becomes clearer to the reader that the broadness of the introduction is because the study’s aim is the use of this tool (random forest algorithm approach) to identify potential factors that could be missed, rather than its use to address any specific hypothesis. I suggest the authors should clarify this from early on as the novel exploratory approach has important benefits that the authors should highlight in order to provide a delineated rationale for their study, which in my view has important benefirts for filtering through complexity in the wild.

 

L193-210/table 1: The authors should explain why the t displacement event model did not take into account number of birds present as a predictor, or the other species frequency and species ID predictors reserved for the role model. Also, there is insufficient justification of why presence was selected for the 2, 5, 10 & 20 minutes prior and not any other period. Was this arbitrary?

 

Table 1 legend: Specify to readers that “what is associated with who is the displacer and displaced” is the role prediction model and “what is associated with displacements” is the event occurrence model.

 

L231-236: Where there any statistical comparisons of species differences in role proportions carried out? This would provide a useful quantification and insight on strategy preference.

 

L257-260: I do not understand this interpretation. If some predictors positively predicted non-displacement feeding, doesn’t it also mean they negatively predicted displacement events? The authors should expand on this here and clarify. Also, following from my earlier comment on why species frequency and species ID were not used in the event occurrence model, from their results here the authors suggest now social environment effects on displacement, but in L262-281 we find that the role of displacer and displaced relies on species id and prior visitation, thus the displacement act does rely on social-environment factors. I suggest the authors should address this, ideally by changing their model predictors or providing solid justifications for the selection and an improved interpretation of results.

 

L312-325: As I mentioned before, I find issues with this interpretation of effects being reserved for physical environment factors. Indeed, the other model identifies that interspecific competition was more common, but species ID factors were excluded from this model.

 

L327-340: These results suggest that interspecific competition is a more common feature in this context than intraspecific contests. Can the authors discuss this?

 

L342-359: Perhaps the authors could discuss resource value versus fighting ability contributions, as comprised in contest theory (Chapin et al. 2019, Behavioral Ecology, Briffa et al., 2020, Animal Behaviour).

Author Response

Please see the attachment

Author Response File: Author Response.docx

Reviewer 3 Report

My comments are in the attached file.

Comments for author File: Comments.pdf

Author Response

Please see the attachment

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

I do not see any clear letter where my criticisms were considered one by one. However....

Reviewer 2 Report

The authors have carried all necessary corrections and in their changes have addressed all my concerns. I would only like to note that in the future, it would be helpful to reviewers if, in addition to the tracked changes, they also indicated in their replies they indicated the lines where changes where applied in response to the reviewers' comments.