Model scale problem

Forums

RSS (Opens New Window)

Francisco Izquierdo, modified 3 Years ago.

Model scale problem

Youngling Posts: 9 Join Date: 10/30/20 Recent Posts

Dear SS experts,

I am fitting a model with simulated data in which all biological parameters (growth, M, etc.) are fixed. The observed fleet selectivity and LFDs are consistent and the SSB trend is similar to the simulated reference one. However, I think there is a problem with the scale of the model.

As I have seen trough different trials, the model can explain the catches with a low F and a high SSB but it could also explain the same catches with a higher F and a lower SSB.

In the current best model, the estimated annual F values are quite low (0-0.06) and the estimated SSB displays huge values (around e+07), while the simulated reference B0 value is around e+06.

Following some pieces of advice in the forum, I have R0 in phase 1 (M and steepness are fixed). In addition, I tried to set a lower phase for DN selectivity slopes (p3 & p4 from 6 to 4) but then the estimated selectivity shapes were not good anymore.

If anyone knows has an idea about how could I solve this problem I would really appreciate it.

Many thanks in advance.

Fran

Richard Methot, modified 3 Years ago.

RE: Model scale problem

Youngling Posts: 222 Join Date: 11/24/14 Recent Posts

Dear Francisco,

What you are finding is not uncommon. The issue is that an assessment model can only infer scale from trend data if there is enough contrast in the data to allow the model to detect the degree to which periods of higher catch have caused steeper declines in the stock abundance indicator (often CPUE). If the model cannot get information about scale from the CPUE trend, then there are other sources of data that can provide scale.

One of these is catchability (Q) if there is a fishery-independent survey that has been calibrated to absolute abundance and is not simply a non-dimensional trend.

Another is age or length composition data, but only if there are strong constraints (or better asymptotic) selectivity for older fish

A third is tag-recapture data that can be informative about Z unless confounded by reporting and tag loss issues.

There are two features in SS3 that allow you to impose a prior on the scale, but be cautious about interpreting the results because the model could simply be responding to the prior and not much influenced by the data.

SS3 also provides two options to impose a prior on scale.

One of these features is "ballpark F" which allows you to tell SS3 a prior on the overall annual F for some selected year.

The other option is "depletion fleet" which allows you to add another survey to the model and designate that survey to be an indicator of the ratio of SSB in that year to the initial year SSB.

Both of these options are described in the user manual.

Rick

On Wed, Apr 27, 2022 at 3:34 AM Francisco Izquierdo <VLab.Notifications@noaa.gov> wrote:

Dear SS experts,

I am fitting a model with simulated data in which all biological parameters (growth, M, etc.) are fixed. The observed fleet selectivity and LFDs are consistent and the SSB trend is similar to the simulated reference one. However, I think there is a problem with the scale of the model.

As I have seen trough different trials, the model can explain the catches with a low F and a high SSB but it could also explain the same catches with a higher F and a lower SSB.

In the current best model, the estimated annual F values are quite low (0-0.06) and the estimated SSB displays huge values (around e+07), while the simulated reference B0 value is around e+06.

Following some pieces of advice in the forum, I have R0 in phase 1 (M and steepness are fixed). In addition, I tried to set a lower phase for DN selectivity slopes (p3 & p4 from 6 to 4) but then the estimated selectivity shapes were not good anymore.

If anyone knows has an idea about how could I solve this problem I would really appreciate it.

Many thanks in advance.

Fran

--
Francisco Izquierdo Stock Synthesis Virtual Lab Forum https://vlab.noaa.gov/web/stock-synthesis/public-forums/-/message_boards/view_message/21935588 VLab.Notifications@noaa.gov

Chantel Wetzel, modified 3 Years ago.

RE: Model scale problem

Youngling Posts: 32 Join Date: 10/7/15 Recent Posts

I agree with the Rick's response on what may be causing this behavior. I would like to add a couple of suggestions that I have found helpful with my own simulation work. I would suggest testing out a perfect data scenario without process and observation error in the simulated data combine with large data quantities for all years within the model. Running this type of model allows you to confirm that the data you are passing to Stock Synthesis is in the correct expected form (e.g. units) for the model. Confirming that you can get estimates that match the simulated population can be really useful as a first check. If you can confirm that everything is set up correctly, then running alternative data scenarios that systematically reduce the data quantity and quality can allow you to better understand the data issues that are resulting in the two alternative model estimates that you are observing.

Francisco Izquierdo, modified 3 Years ago.

RE: Model scale problem

Youngling Posts: 9 Join Date: 10/30/20 Recent Posts

Dear Rick,

Thank you for your explanation and advice on how to solve the problem. I applied them as described in the manual and I think they are very useful. However, the "F balpark" didn't work and regarding the "depletion survey", you have to be sure you know the ratio between B0 and SSB0 for that year as you somehow force SS to follow it.

After some trials I think that the problem on the scale may also come from the single area configuration. The dataset has been simulated for a spatial model but I am still fitting everything to a single area model first, so probably, when I increase the number of areas, the tag data will be more informative on the model scale as you mentioned.

Thanks also to Chantel for the suggestions although in this case I have not been the one to simulate the data, so I don't have all the information about quantities that I would need.

Regards,

Fran