A better uncertainty estimate of total seabird bycatch
Assessing the uncertainty of total seabird bycatch synthesized from multiple sources by Zhou and Liao
Background The highly migratory behavior of seabirds has implications for the uncertainty calculations of the total seabird bycatch estimate at a regional/global scale synthesized from individual assessments conducted at a local scale.
Problem statement To efficiently characterize the accuracy and reliability of fisheries direct impact on seabird populations.
Contribution In addition to a theoretical exploration, we provide a hypothetical scenario analysis based on data from the Western and Central Pacific Fisheries Commission convention area. The results show that the assumptions on the correlation between different areas has a big impact on the uncertainty estimates, especially when the number of areas to synthesize is large, and simplifying assumptions failed to capture the complex dynamics of seabird bycatch rates among different areas.
Recommendations Empirically estimate the correlation of bycatch rates between each pair of sources when time series of bycatch rates are available.
The central idea of this study can be summarized in one sentence - the traditional approach of tallying total seabird bycatch from individual assessment reports over-estimate the uncertainty bounds by A LOT! It turns out that we can do much better using the strategy detailed in this paper. The meat of this study is in fact the supplement hiding at the tail section of this article. One of the reviewers feared that those equations would turn away many potential readers who actually may benefit from reading it (my own interpretation). The editor suggested and we all agreed that hiding those equations in a supplement would be a better approach. Those who need those formulae can still find them, and those who don't want to see them won't. The rest of the article provides nessesary biological context/background, and a solved hypothetical problem from the Western and North Pacific. The strategy detailed here works not only for seabirds, but also for other highly migratory species, like marine mammals, 🦈 🤷 and turtles.
Some further discussion: Most current and past analyses/reviews on a regional/global scale are based on summary statistics, such as mean and standard error. It is like this because of the confidentiality concerns of the raw data sets. Meanwhile, there exist some notable efforts to pool all the raw data into a single dataset and perform the reginoal assessment as if doing a local assessment. If somehow you have convinced everybody to surrender their data, there are some further disadvantages of that approach, including 1) it homogenizes all the datasets such that the emphasis is on some global effects and not local effects (meaning that area specific estimates will be biased), 2) areas with large sample sizes most likely will unduely influence areas with smaller sample sizes, and 3) it repeats what local assessment is doing but in an inferior way.
Just think about this for a moment. What we need from each area is an unbiased estimate of the number of seabirds killed in that year. Nothing more and nothing less. Given that estimate, we don't care what your observed sample look like or which method you use, provided they are scientifically sound. In our opinion, each individual assessment is better done at the local level using the method that suits the local situation the best, and at the higher level, only those summary statistics will be needed. Voila! No awkward data requests or unnessesary confidentiality agreements. Everyone is happy.
Let's say you are still not convinced. I'll show you a bonus picture then (see below ⬇️, not above). Let's do a thought experiment. Say you are doing a paired experiment at two locations. In this experiment, whatever you do at one location, you replicate at the other location simultaneously. Now, you are collecting data on seabird bycatch rates at these locations. What you have are pairs of bycatch rates. You plot these pairs in a scatter plot and now you have part of the left panel below (A): those points fall in that grey shaded area. You point cloud may tilt either to the left or to the right, and the exact slope is not important for our argument here. I have to draw it in some definite manner. For now, let's keep it sloping upwards.
Now, just pause for a moment, and consider the real situation. In reality, you don't have that paired experiment, and you can't plot those imaginary pairs in A, since you only have singletons with missing data from either one or the other location. What you have is a marginalized view of what you could have in that parallel universe, I mean the paired experiment. What you have now are those univariate distributions from either location (curves on x and y axes in A). In order to calculate the total seabird bycatch, you need to know the value of that slope. The marginalized view doesn't have the slope information, and you may have any slope that is consistent with your marginalized view (B), but only one of those infinite choices will produce the correct answer. The core of our approach is to replace that missing slope with a substitute. And we chose paired annual bycatch rates. Now you may read the method section for more. There is no point in repeating what is already there.
I hope I have provided enough motivation for you to download the full paper and read it yourself. It is open access and you don't have to pay a quid to read it. Here is the link:
10.3390/birds3030017
Cite this paper if you like it ❤️
or leave us a comment 👇