Sablefish_Reference_Analysis/MREF/README.md

In John-R-Wallace-NOAA/FishNIRS: NIRS on fish structures.

For quality control, four reference otoliths from three different species are scanned in the morning before production scanning commences. This is done regardless of the species being scanned that day (or session). These are the "MREF" scans. More MREF scans may also be done throughout the day if there are issues that needs investigating. Sablefish, with their relatively small sized otoliths is one of the three MREF species. Sablefish otolith reference scans are analyzed below since the comparisons of the daily reference scans to their respective baseline scans only changed for those otoliths.

A large Fully Connected Neural Net (FCNN) model with both spectra scans and metadata from 5,235 Sablefish otoliths collected on the 2017, 2018, 2019, and 2021 NWFSC Bottom Trawl Surveys (All_Oties_2017_18_19_21) was used to predict the age of the four Sablefish reference otoliths. The metadata consisted of otolith weight, fish length, fish weight, and depth of capture. This agreement figure shows the prediction of ages vs the traditional method of aging (TMA) for all the otoliths in the model and for the 1,553 otoliths from the 2022 survey, which were not included in the 'All_Oties_2017_18_19_21' model (5,235 + 1,553 = 6,788).

The four Sablefish reference otoliths have TMA ages of 2, 18, 38, and 51. The FCNN model was used to predict ages from each daily scan which, subsequently, were plotted against a timeline of days since first use of the Bruker Tango in Newport, OR. (Not every day is a scanning day, of course.) This figure includes the month interval labels and also the mean and standard deviation of the NN model predicted ages.

Another FCNN model, utilizing 500 random selected otoliths from each of the 2017, 2018, 2019, and 2022 NWFSC Bottom Trawl Surveys, plus an additional 200 randomly selected from 2021 (500N_2017_18_19_22_200N), was also used to predict the four reference otoliths in the same manner.

Before the otolith weights of the MREF Sablefish were obtained, a 2022 (survey) spectra only model was applied to the four otoliths. (Forty three scans were left out of this FCNN model to check for prediction bias, and the number of predicted scans is slightly higher at 1,556, since no otoliths were removed due to missing metadata.)

Lastly, since the four Sablefish reference otoliths are from 2019, a spectra/meta model from only the 2019 survey was also applied to the four otoliths. Unfortunately, there are only a limited number of older Sablefish in 2019 for the model to work with, and the total number, reduced by missing metadata, only comes to 750. (The total number of scans is 802.) However, it is instructive to include it here.

A number of points can be drawn from these figures: - For the large 'All_Oties_2017_18_19_21' model the Sablefish reference otoliths with a TMA of 38 and 51 are very well predicted over time. However, the otoliths with a TMA age of 18 and 2 are consistently off by ~2.5 and 1 year(s), respectively. A traditional re-reading of those bones may prove insightful.

• The '500N_2017_18_19_22_200N' model also shows the Sablefish reference otolith with a TMA of 38 well predicted over time, and again the otoliths with a TMA age of 18 and 2 are consistently off by 2.5 and 1 year(s).

• The spectra/meta model of 2019 and the spectra only model of 2022 more easily show that the reference scans settle down over time, around March. This settling can also be seen in the All_Oties_2017_18_19_21 and 500N_2017_18_19_22_200N models, however that difference cannot be visually detected after around January. One idea for the settling is a reduction of shininess, and hence lower reflectiveness, of the bottom of what was originally a brand new "gold stamp" that is used to cover the otoliths during scanning. The Tango Technicians may have better ideas.

• Note also, that the predictions can be seen to slightly increase at the very end of the time series for all the models except the spectra/meta model of 2019. This may be the change in the scans that prompted this analysis and was perhaps caused by the Tango being over its operating temperature of 35 degrees C during midsummer. Due to GitHub controls, this Tango spec sheet URL needs to be copied and pasted to work: < https://file.yzimgs.com/427772/2013092310065343.pdf >

John-R-Wallace-NOAA/FishNIRS documentation built on April 12, 2025, 12:59 a.m.