Advice on length-based selectivity

[Laura Lee]

Hi all,

My current model includes a fisheries-independent gill-net survey (FCI_gillnet) that catches a broad range of sizes (see attached). I have performed runs that assume a double normal selectivity (pattern #24) and runs that assume a cubic spline (pattern #27) over a range of number of nodes. The sample sizes for the available length data for this survey are low so I am using a super-year approach to characterize the lengths. All my experimental model runs result in poor fits to the length data for this survey (see attached) and predicted selectivity curves that don't seem reasonable (see attached). I'm wondering if another selectivity function would be worth investigating or perhaps my effective sample size for the super-year is too low to expect a decent fit (Nsamp = 4.2).

I would appreciate any advice as to how to move forward. Thanks in advance for your time.

Best,

Laura

P.S. The attached figures are from a run that assumes a cubic spline selectivity with 4 nodes.

[Jason Cope]

Hi Laura,

You are correct that the very small sample size does not encourage the model to care much about those lengths.

I am curious how large the gillnet removals are compared to other catches (i.e., how important modelling that selectivity will matter in the model).

If you want to send me your model files (just the data, control, forecast, and starter files), I can take a look and see if I see anything that may improve the fit or help explain what is going on.

-Jason

[Michael Schirripa]

Laura;

One thing you might try first is to combine your sexes for this survey. That would at least increase your sample size. If that doesn't help as much as you like then you may want to consider how you are weighting these lengths comps. Are you using re-weighting techniques? ALso, I would ask, how much is your result changing by achieving a better fit to these lengths? If the results are not changing much when you increase the weighting then you might have to be so concerned that you are not fitting a length comp with a small sample size. What is pushing against the better fit? That is, if you remove the lengths entirely which of the other data fits better? This is all to say, there are other ways to look at this rather than just changing selex models.

Mj

--
You received this message because you are subscribed to the Google Groups "SS3 - Forum" group.
To unsubscribe from this group and stop receiving emails from it, send an email to ss3-forum+...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/ss3-forum/2dbd66b1-5dae-4e2f-9432-7511a353e56fn%40googlegroups.com.

[Mark Maunder]

Hi Laura,

 

See section 3.6 in this document for our current thinking on selectivity

 

https://www.iattc.org/GetAttachment/11b8757d-ad0f-435c-b2c6-40f5658c56c0/SAC-16-03_Yellowfin-benchmark-assessment---2025.pdf

 

I did not see the plots and don’t know exactly how the length comps were created, so can’t comment much. But also wanted to add that spatial weighting of the length comps by CPUE is generally better than simply using raw comps or catch weighting.

 

I hope this helps,

 

Mark

--

You received this message because you are subscribed to the Google Groups "SS3 - Forum" group.
To unsubscribe from this group and stop receiving emails from it, send an email to ss3-forum+...@googlegroups.com.

To view this discussion visit https://groups.google.com/d/msgid/ss3-forum/534a18b4-6086-4060-b34a-3c4ba3183f74n%40googlegroups.com.

[Mark Maunder]

Hi Laura,

 

I just received your email with the plots. Something is definitely messed up. Try a run with the input sample size set at 200 or something and see if it works better. If not, then there is a setting that is off related to the smallest size fish that is wrong.

 

I also don’t particularly like the super year approach.

 

Given the low sample size, if you think the index of abundance is reliable, you probably should estimate the selectivity with a high weight on the comp data, then fix the selectivity at that estimate and rerun the model without fitting to the comp data. This assumes that there is a lot of data for the index but not much data for the comps.

 

Regards,

 

Mark

 

From: ss3-...@googlegroups.com <ss3-...@googlegroups.com> On Behalf Of Laura Lee

Sent: Friday, September 12, 2025 8:51 AM
To: SS3 - Forum <ss3-...@googlegroups.com>

--

You received this message because you are subscribed to the Google Groups "SS3 - Forum" group.
To unsubscribe from this group and stop receiving emails from it, send an email to ss3-forum+...@googlegroups.com.

To view this discussion visit https://groups.google.com/d/msgid/ss3-forum/2dbd66b1-5dae-4e2f-9432-7511a353e56fn%40googlegroups.com.

[Ian Taylor - NOAA Federal]

Hi Laura,

One more thought on top of the good ideas that the others shared: I think the spline fit in your original attachments may not be as bad as it looks. If there are very few fish in the population below 8cm then the strange spike at the start of the curve won't matter. But the blue and red lines in the comp fit plot are scaled as proportions so those spikes make it really hard to see the shape of rest of the curve, which might be a not-terrible match to the observations. I wish I had better guidance on choosing knots for the spline, but if you want to continue down that path, I would start with reducing to three knots and testing alternative locations for them to try to avoid the spike for the smallest bins.

-Ian

To view this discussion visit https://groups.google.com/d/msgid/ss3-forum/SN6PR02MB5648D2538AB37118258A56E1CC08A%40SN6PR02MB5648.namprd02.prod.outlook.com.

[Laura Lee]

Thanks everyone for the feedback! Lots of ideas to investigate and I am very appreciative.

Best,

Laura