respecify length bins

[sgaichas]

We started with equally spaced length bins throughout the range of observed lengths across fishery and survey data for each species. This made inefficient use of data (smallest and largest bins often missing data).

Options discussed in January 2023 include

• Start from min observed, not 0 for bin 1
• Combine bins 1-2? Finer structure thereafter? Bigger last bin? Plus group

A new bin definition algorithm implemented in hydradata first calculates quantiles for each species based on all input lengths aggregated over time. The current implementation uses the smaller of the survey or fishery 10%ile as the minimum size for bin width definition, and the larger of the survey or fishery 90%ile as the maximum size for bin width definition. Equal bin widths within this range are calculated, and then the first and last bin are extended to include 0 for the smallest and the max observed length for the largest bin.

A visualization of bin definitions (black vertical lines) for each species and aggregate dataset is below, based on the current (January 2023) mskeyrun Georges Bank dataset and 5 length bins:

Thoughts?

[sgaichas]

implemented in new datasets called hydra_sim_GB_new5bin_1978_10F commit f62fc60

hydradata commit

[sgaichas]

apologies, latest dat file push now runs with hydra-sim.tpl currently in main branch. I neglected to remove the otherfoodterm that you have been commenting out by hand until now.

[sgaichas]

Diagnostics including raw length comps divided into new bins from a test run "estimate everything": posted here

[sgaichas]

@gavinfay when I make the new dat file with the 3 fleet inputs (and incorporating vulnerability), do you want the newer length bins shown here (in this issue) with wider "tails" or do you want the equal spacing bins? The current (May 2023) dat file you posted that adds vulnerability has the original equal spaced bins.

[gavinfay]

I think the wider tails make sense to me. Thanks!

…

On Wed, Jun 7, 2023, 3:40 PM Sarah Gaichas ***@***.***> wrote: @gavinfay <https://github.com/gavinfay> when I make the new dat file with the 3 fleet inputs (and incorporating vulnerability), do you want the newer length bins shown here with wider "tails" or do you want the equal spacing bins? The current (May 2023) dat file you posted that adds vulnerability has the original equal spaced bins. — Reply to this email directly, view it on GitHub <#31 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AB5X65ANUQWTVMON44IK2JLXKDKJ7ANCNFSM6AAAAAATYTIMWY> . You are receiving this because you were mentioned.Message ID: ***@***.***>

[sgaichas]

new bins are added in 7550767

[sgaichas]

Update to survey lengths in mskeyrun results in these new aggregate length distributions

Corrections to survey length data slightly changed the bin widths based on the rules determining them,
time series of proportion in length bins updated. new dat and -ts.dat files in 883fd3f

the model runs with new input files.

[gavinfay]

Excellent. Thanks Sarah!
Those are some teeny cod caught in the survey :)
Odd that the 1st and 2nd length bin dividers don't seem to match the '10th percentile' rule, for quite a few of these (e.g. haddock, herring, mackerel, silver hake). Seems particularly odd for the haddock distribution (almost all the survey lengths are in the first bin). I know that the length bin structure is also taking into account fishery lengths so maybe that is influencing this.

[sgaichas]

Yes, correct, the bins are defined based on both the fishery and the survey lengths combined distribution. Here are the fishery data for comparison:

So it could be this isn't the best approach for determining bins given the differences in the two datasets? I'm happy to explore alternatives.