Atlantis to SS comparisons

Forums

RSS (Opens New Window)

Bai Li, modified 2 Years ago.

Atlantis to SS comparisons

Youngling Posts: 3 Join Date: 10/9/19 Recent Posts

Hello SS team,

I am working on setting up a stock synthesis model using simulated data from Atlantis operating model. The case species is Norwegian Barents cod. Three cases have been tried:

1) atlantis2ss_full_data that uses fishery and survey age composition data. Model had strong patterns in age composition residuals and large final gradient value of 6.88787e+08

2) atlantis2ss_surveyagecomp that uses survey age composition data. Model still did not fit well.

3) atlantis2ss that uses fishery age composition data but not survey age composition data. Model had a better final gradient value and fit.

However, the estimates from SS and true values from Atlantis did not match sometimes. I wonder if you have any thoughts on the discrepancy and suggestions for modifying some of the input values. All comparison figures from case 3 (atlantis2ss) can be found in this README.MD.

We are still investigating if the Atlantis age comp output and recruitment are behaving as expected. Thanks for your help!

Best,

Bai

Ian Taylor, modified 2 Years ago.

RE: Atlantis to SS comparisons

Youngling Posts: 117 Join Date: 12/8/14 Recent Posts

Hi Bai,

There's a lot to digest here, but I took a quick look at the atlantis2ss_full_data model. My sense is that the model estimation is getting stuck in a state where the population is crashed, the catch penalty is invoked from not being able to remove the full catch, and small changes in parameters have little impact.

You might try changing the phasing so that log(R0) starts large and is only estimated in a later phase after selectivity has been better estimated. You could also explore the "Ballpark F" option (User Manual description), which can be used in the early phases to keep F at a reasonable value.

I'm also a little worried about the estimability of the non-parametric (type 17) selectivity for the census fleet with so many ages, but the data seem abundant enough that it may work if the scale issue can be resolved.

-Ian

Richard Methot, modified 2 Years ago.

RE: Atlantis to SS comparisons

Youngling Posts: 219 Join Date: 11/24/14 Recent Posts

Thanks Ian.

Even better (I think) than ballpark F, is depletion fleet which is essentially ballpark degree of stock depletion, which is essentially DB-SRA. This also is described in the manual. What it does is add another survey fleet that takes SSB as its input. There is an option to implement it in a way that inserts a new phase 1 in which only ln(R0) is estimated. So if target depletion is set to 0.5 in the endyr, SS3 will conduct a first phase in which lambda=0 for all data expect the depletion fleet, then SS3 quickly finds the ln(R0) that will match that degree of depletion, then proceeds to subsequent phases where other parameters get estimated and all data types get their assigned lambda. I think this feature should routinely be used in all SS3 configurations.

Rick

Richard D. Methot Jr. Ph.D.
NOAA Fisheries Senior Scientist for Stock Assessments

Mobile: 301-787-0241

On Thu, Dec 9, 2021 at 11:34 AM Ian Taylor <VLab.Notifications@noaa.gov> wrote:

Hi Bai,

There's a lot to digest here, but I took a quick look at the atlantis2ss_full_data model. My sense is that the model estimation is getting stuck in a state where the population is crashed, the catch penalty is invoked from not being able to remove the full catch, and small changes in parameters have little impact.

You might try changing the phasing so that log(R0) starts large and is only estimated in a later phase after selectivity has been better estimated. You could also explore the "Ballpark F" option (User Manual description), which can be used in the early phases to keep F at a reasonable value.

I'm also a little worried about the estimability of the non-parametric (type 17) selectivity for the census fleet with so many ages, but the data seem abundant enough that it may work if the scale issue can be resolved.

-Ian

--
Ian Taylor Stock Synthesis Virtual Lab Forum https://vlab.noaa.gov/web/stock-synthesis/public-forums/-/message_boards/view_message/19366379 VLab.Notifications@noaa.gov

Bai Li, modified 2 Years ago.

RE: Atlantis to SS comparisons

Youngling Posts: 3 Join Date: 10/9/19 Recent Posts

Thanks Ian and Rick for your help. The scale issue is getting better after adding a depletion fleet. Will work with Atlantis folks to check input data and then fine tune the SS3 model again. -Bai

Richard Methot, modified 2 Years ago.

RE: Atlantis to SS comparisons

Youngling Posts: 219 Join Date: 11/24/14 Recent Posts

That's good to hear.

I wouldn't get too hopeful. The data show near constant catch for the entire time series and the survey also shows little trend. So little surprise that scale is hard to get right.

Given that your model has no initial equilibrium catch (catch with year = -999), the SS3 model time series is assuming that the population is virgin in the start year (40).

Then with no trend in the index data, SS3 is getting all scale info from the composition data. But the composition data show 70% of the fish in the first age bin, beginning in 40. So there is no contrast here either. It seems the Atlantis is generating a population that is already highly depleted by year 40, then SS3 starts in year 40 and is told that the population is virgin going into year 40.

I'm 99% certain that this is the source of the mismatch.

Rick

On Fri, Dec 10, 2021 at 2:18 PM Bai Li <VLab.Notifications@noaa.gov> wrote:

Thanks Ian and Rick for your help. The scale issue is getting better after adding a depletion fleet. Will work with Atlantis folks to check input data and then fine tune the SS3 model again. -Bai

--
Bai Li Stock Synthesis Virtual Lab Forum https://vlab.noaa.gov/web/stock-synthesis/public-forums/-/message_boards/view_message/19392018 VLab.Notifications@noaa.gov

Bai Li, modified 2 Years ago.

RE: Atlantis to SS comparisons

Youngling Posts: 3 Join Date: 10/9/19 Recent Posts

Hi Rick,

Thanks for your advice and I will share your insights with the folks from the project.

Best,

Bai