Outside the box, on a tangent about fish
Climate change has a variety of facets, and often we scientists become somewhat pigeon-holed in our viewpoints and disciplines. Sometimes there is much more to be gained by tackling out of field problems, and this is how we now stumble upon the subject of fish, and, in particular, salmon.
Salmon scales, as it happens, grow in a somewhat similar fashion to tree rings, with circuli that become more spaced as the fish grows faster, and narrower otherwise. Biologists use these patterns of fast (summer time)/slow (winter time) growth to identify the age of the fish once it is captured and released. However, the commonly implemented way to determine these patters is by hand, with a human operator manually selecting an arbitrarily "longest" transect of the scale and counting the circuli spacing patterns by hand. Furthermore, this approach is plagued by measurement noise, rendering bulk analysis of subtle growth rate patterns intractable.
A scale can however be considered as a series of radially distributed almost-harmonic time series, lending it to Fourier analysis.
As such, in collaboration with Fisheries and Oceans Canada, we developed a semi-automated framework with a publicly released browser architecture (https://scalefyer.jchaput82.workers.dev/) that allows biologists to process not only large amounts of fish scales efficiently, but permits averaging over the entire scale rather than just a single transect.
As a direct port from traditional seismic methods, we process each transect of the scale from center to edge and assemble spectrograms. Since the spectral information of each transect is nearly harmonic, we can "stretch" the spectrograms for each transect and average them over the entire scale, resulting in a very clean "growth pattern" for that fish. Importantly, the robustness of the extracted information has permitted, for the first time, the systematic identification of fluctuations in growth rates endemic to all fish of that year that vary significantly with years. This implies that salmon during their somewhat cryptic ocean dwelling periods are generally subject to global conditions rather than local ones, pointing to strong climate-based forcing on salmon populations. A formal study of over two decades of Atlantic salmon populations is now underway as compared with NOAA sea surface temperatures and other available metrics of climate forcing.
On the left is an example of 50 scales for three different years of salmon captured on the Miramichi river, Canada, processed with our method (left) vs human operator (right). Beyond significantly reducing errors, we notice that each row representing the growth rate of a fish of a given year in the top left panel is more coherent with its cohort (i.e., the spectral growth patterns among fish of the same year are nearly identical during the summer ocean period) and those patterns vary drastically between years (middle panels).
A proposal for a more systematic study using legacy datasets from Norway and Denmark is being discussed, and our software is now available to the community in browser form at no cost.
(https://scalefyer.jchaput82.workers.dev/)
This project is active, and will be looking for student participation in the near future.