noaa-afsc / berchukseals-haulout Goto Github PK

It's not explicitly stated that the 'second' model fit included all of the same covariates and interactions as the first model fit. The only addition is the inclusion of the year:day and year:day2 interactions.

The findings are appear robust and intuitive. The authors need to improve the presentation of the findings to improve clarity and help the reader understand how they addressed the knowledge gap.

Checklist

• Addressed in paper
• Addressed in reply letter

After some discussion and exploration, we're considering updates to the model specification with a few key differences:

1. 4th degree polynomial for day of year (day)
2. removal of interactions with sin1, cos1, etc

The 4th degree poly allows for some additional flexibility such that the period of time for adult females when they are nursing is represented better. The 3rd degree polynomial seemed to have smoothed over this period too much.

The removal of interactions was done to limit the complexity of the model and allow those features to be captured within the temporal autocorrelation and not as part of the mean. Generally trying to limit terms to those for which we can attribute a biological/ecological meaning.

I've identified some data inconsistencies and data not included in the analysis that should be included. The main reason for this seems tied to the deployments not having a specified 'end date' in our internal telemetry database.

At a minimum, the issue impacts the following spenos:
PL2019_9001,PL2019_9002,PL2019_9003,PL2019_9004,PL2019_9005,PL2019_9006,
EB2019_9001

Will work with @staciekoslovsky-noaa to resolve in the database and proceed with additional data integrity tests.

Reviewer 1 comments:

• "Please use standard figure titling and not use a statement that should below in a PowerPoint slide or in your Results and Discussion section."
• "Your plot indicates a y-axis of local solar hour which varies from 0 to 24.  This is clearly not fixed at local noon."

Checklist

• Addressed in paper
• Addressed in reply letter

Many data components of this analysis rely on a pull from an internal PostgreSQL database. This has two implications:

1. the data on the database can change/update and impact downstream analysis and results along with future reproducibility
2. there's no option to version control the data

To resolve this, the plan is to create local data stores within the repository as either *.csv or *.parquet files

@pconn comment:

I continue to think there are better options for this plot, even with the changes. It just seems like it’s unnecessarily large, and that there isn’t any compelling reason to present days vertically. To my mind, a simple line plot with date on the x-axis and seal-hours on the y-axis (with a line for each species) would be a lot simpler to interpret and would take up less space.

The more I work with this, the more I agree. I have some ideas and will re-work and share those options here

reference this article:
https://www.nature.com/scitable/topicpage/effective-writing-13815989/

see comments and example from J Crawford

The initial attempts to plot the marginal/conditional effect of individual weather covariates on haul-out probability resulted in plots that were aesthetically pleasing and 'beautiful' but, ultimately, distracting and difficult for the reader to interpret.

here's example of the initial plot for spotted seals

the following collection of reviewer comments from Jason Baker sums up the rationale for improvement

The swooshy vapor trails, I think, require more explanation. You mention transparent vertical lines, and I can see some of those, but I can also see horizontal bands in some places. How to interpret the alternating consecutive high and low CI’s on the right 2/3 of the precip graph, which mostly don’t include the fitted line!?

It comes down to what you want to convey. If it’s that we need to account for temperature, then what is wrong with good old marginal plots? You could choose peak haul out, high solar noon, mean precipitation, mean wind, etc. and show how temperature affects haul out probability.

In chatting with @dsjohnson I'm convinced that traditional marginal or conditional effects plots aren't the best option for communicating the relationship. These plots pull from the same data set as the haul-out probability surface plots and, simply, reflect the range of predicted haul-out probabilities across the range of weather covariate values. So, I also don't have to develop all the additional code to produce more typical marginal/conditional effects.

That said, I think Jason's concerns are important to recognize. I'm hoping that I can find a solution by limiting the range of days and hours in the plot and provide some control for underlying temporal collinearity and interactions that lead to the 'swooshy vapor trails'

comment from Lori Q and others suggests improvements to the description of weather covariate effects. For example, ribbon seal section provides more details on whether impact was positive/negative. But, bearded seal section has minimal detail.

The modeling efforts are rigorous and robust, improving on previous modeling work. I still had one remaining question on why the authors couldn't run the models with and without the sea ice concentration as a predictor. The comparative results might be interesting (I understand why the correction factor might be biased...but given that this is a baseline study, it would have been an interesting comparison to see whether it was or wasn't a decent predictor for the different seal species).

Checklist

• Addressed in paper
• Addressed in reply letter

From PLB re: potentially examining trend of date of peak haul-out prob vs year

Another trend that may be of interest is a trend in date of peak haul-out prob vs year. Because the colored symbols of the peaks don’t follow a color ramp, it’s really hard (at least for me) to judge whether there could be a trend. Another way to look at it would just be a scatter plot of the dates vs year?

The authors provide a throughout treatment of a rich dataset acquired across numerous studies and united here into a single modeling effort to improve the ice seal abundance estimation. This is valuable work and merits publication.

The authors have put substantial efforts into setting the context for their analysis, citing relevant studies. I would like to see a more direct approach in the framing of their problem: the improvements of abundance estimates required by the Marine Mammal Protection Act and Endangered Species Act. I would like to see a specific example of how the modeled availability would have affect population abundance estimates by apply the availability model to aerial survey data and providing an adjusted abundance estimate.

Checklist

• Addressed in paper
• Addressed in reply letter

The authors need to explain more how they are attributing statistical significance. For example, in lines 344-345, they ascribe strong influence of temperature with a p-value of 0.064.

Not to be one who is fixated on a p-value of 0.05 denoting significance, I was a bit surprised to see the same authors, in lines 364-366, to state that there is 'no indication that the observed trend is meaningful'for spotted seal adult females having an R2 of 0.767, p=0.022. If the extent of sea ice explains nearly 80% of the variance in peak haul-out probability, how is it that the author's call this a 'trend'? This requires more explanation. As does the statement in line 427-428.

Checklist

• Addressed in paper
• Addressed in reply letter

The computed manuscript approach outlined in this repository has a lot overlap with the research compendium approach used here.

Will look for opportunities to borrow from the computed manuscript approach and enhance this repository.

We use 'environmental covariates', 'weather covariates', and 'meteorological covariates' to represent the same set of covariates throughout the paper. Should settle on either 'weather' or 'meteorological' ... I think 'environmental' is too broad when, here, we're specifically focused on covariates from NARR.

Maybe check the walrus manuscripts for their terms

The color ramp addresses previous comments about no being able to differentiate lower percent dry values (i.e. you can, now, see some of the variation when seals are predominantly in the water). But, the pale yellow at the high end makes it difficult to see the empty areas since the background is white.

I've tried adding additional markers to the plot to specifically call out the empty areas, but the results were not satisfactory. Might be best to worry less about the lower end variation since a main purpose of the figure is to indicate the presence of missing data.

One suggestion I would share with the authors is the addition of a limitations section in the discussion. The limitations were outlined throughout the methods, but a succinct accounting of these limitations in the discussion would allow readers who do not want to carefully dissect all of the modeling in the methods to be able to better interpret the robustness of the results and conclusions.

Checklist

• Addressed in paper
• Addressed in reply letter

@pconn suggests we consider removing the following paragraph from the introduction or including some of the key points within the Discussion section

Ultimately, knowledge of trends in phenology and abundance (or life history
surrogates such as survival and recruitment) will be necessary to make credible
quantitative predictions about the effects of climate disruptions on the
abundance and distribution of Arctic seal populations. Before we can construct a
trend, however, we first require a baseline. Several studies have contributed
estimates of the distribution and abundance of ice-associated seal species in
the Arctic using aerial surveys (e.g., [@bengtson2005a], [@conn2014a], and
[@verhoef2014a]). Such abundance studies were conducted over very large areas and
estimation of absolute abundance required making inference about numerous issues
affecting the observation of seals on ice. These included availability (only
seals basking on ice were available to be counted), detection probability
(observers or automated detection systems may have missed some seals on ice),
species misclassification, and possible disturbance of seals by aircraft
[@conn2014a; @verhoef2014a]. Refining these inferences will improve the
accuracy of abundance estimates in the Arctic.

If we decide to submit to PeerJ instead of Royal Society (see #13), then there will need to be some editing and reformating of the manuscript

The current word count needs to be trimmed by ~3000 words and 4 figures to meet the page limits for Royal Society.

The manuscript currently states that the response variable is hourly percent dry. This is not correct.

create_model_input() rounds the percent_dry value to either 0 or 1
https://github.com/jmlondon/berchukHaulout/blob/8a29f9e7feea078298e923d1ff450d91d7796ac9/R/create_model_input.R#L3

• update text and clear up any potential confusion regarding the nature of the response variable
• add references to London et al and other recent studies that use Bernoulli
• calculate the percentage of observations below 10% and above 90% and report

from @JuCrawford:

" ... you had mentioned Barrow/Utqiagvik was the northern/eastern most location seals were tagged. I thought I had sent data for seals tagged in Nuiqsut also"

PeerJ Staff Note: Please ensure that all review and editorial comments are addressed in a response letter and any edits or clarifications mentioned in the letter are also inserted into the revised manuscript where appropriate.

Checklist

• Addressed in paper
• Addressed in reply letter

In the Introduction, the authors should indicate the location of where previous studies on seals had taken place.

Checklist

• Addressed in paper
• Addressed in reply letter

@pconn comment:

Given that there isn’t a figure, it might be best to omit the bit about “trend line” – and to provide more explanation. For example, “Adult female and subadult spotted seals appeared to achieve earlier peak haul-out date in years when ice concentration was high” (or whatever it was).

This paragraph was originally written when there was a figure. So, probably worth considering rewrites like this and maybe a few others. Will look into it.

Either upload a new version with the appropriate file name or change the URL

the Data Availability section could benefit from the addition of direct links to Movebank and US ATN repositories.

suggestion from @JuCrawford to add mention to the bearded seal results

The intent of figure 1 is to show the distribution of raw behavior observations from March through July. I went with a calendar-like layout with weeks in columns because I thought that might be more familiar for readers. Comments from Justin (and Lori) suggest this figure could be improved.

Justin suggested fixing the upper left square to the start day-of-year (i.e. March 1) and extending the columns to represent 10 days (3 columns in a month). I think this is doable within the general framework of the figure code .. and would remove the somewhat awkward inclusion of the weekday row labels.

I think, maybe, 8 days per column would be a nicer spread. Will need to think how best to label the rows (01-Mar, 02-Mar, .. ?)

I found 463-464 confusing. This is an important part of the discussion (using your framework to assess previous results from another study), but it could be more clearly written. The sentence, 'Applying models that ignore age, sex, and year effects...under the current analysis framework' is very confusing and doesn't seem to follow the argument of the paper easily. Could this be reworded so that the full intent of the paragraph is communicated?

Checklist

• Addressed in paper
• Addressed in reply letter

Please check the possible typo in line 263...'only -99% of our observations fell...'

Checklist

• Addressed in paper
• Addressed in reply letter

The following paragraph starts line ~1338

Predictions of absolute haul-out probability in this paper were somewhat
different than those previously reported for these species, especially for
bearded seals. For instance, Ver Hoef et al. [-@verhoef2014a)] and Conn et al.
[-@conn2014a] used haul-out correction factors with maximums of 0.66 for bearded
seals, 0.62 for ribbon seals, and 0.54 for spotted seals, where maximums
corresponded to times near solar noon in mid-late May. Applying models that
ignore age, sex, and year effects, these probabilities were 0.38, 0.72, and
0.60, respectively, under the current analysis framework. Our current estimates
estimates of haul-out probability reflect increased sample sizes in terms of
number of animals, but also improvements to the way data were prepared prior to
analysis.

The values of 0.38, 0.72, and 0.60 need to be re-calculated with the current models.

I appreciate the acknowledgement that it is difficult to know whether differences in abundance estimates are attributable to changes in abundance or changes in haul-out behavior and the potential proposed solution (lines 450-458) Has anyone done a smaller-scale study that would address this inherent problem?

Checklist

• Addressed in paper
• Addressed in reply letter