Recruitment bias in the forecast

[Massimiliano Cardinale]

Dear All,

 

Agurtzane and myself have been discussing how to estimate recruitment in the forecast when the bias correction option is applied to the recruitment deviates and when the SR is used. In SS3, there are 2 options 1) to don’t apply the correction or 2) apply the correction of 1, which is also applied in the historical part, so that it is not possible to use the parameterization suggested by SS3, which is considered as good practice. In the paper about the bias correction as well as in the SS3 manual, I understand that the recommended option is to not apply the bias correction to the forecast recruitment. However, this could imply a mismatch between the mean historical recruitment and the forecast. Could you explain us what is the recommended option and why? In tuna assessment we usually do projections for 10 years with SS3 and these assumptions could impact the results. However, the manual also says when MCMC are run then the recruitment correction should be applied. Why is that? Do they converge towards the same value with the maximum likelihood and without correction in comparison to MCMC runs with corrected option?

 

Cheers  

 

 

“I had never been salaried and felt that the research should be its own reward”

Gorillas in the Mist

 

Dian Fossey

 

 

Massimiliano Cardinale
Associate Professor

Sveriges lantbruksuniversitet
Swedish University of Agricultural Sciences

Department of Aquatic Resources (SLU Aqua)
Institute of Marine Research
Turistgatan 5, 45 330 Lysekil
Phone: +46 761 268 005
massimilian...@slu.se www.slu.se/akvatiskaresurser
SLU Aqua on Facebook

Our vision: Viable fish stocks in healthy waters

Please consider the environment before printing this e-mail

 

 

---
När du skickar e-post till SLU så innebär detta att SLU behandlar dina personuppgifter. För att läsa mer om hur detta går till, klicka här
E-mailing SLU will result in SLU processing your personal data. For more information on how this is done, click here

[Richard Methot - NOAA Federal]

Hello Max and Agurtzane,

Thanks for asking this question.  The concept of bias correction has been difficult to communicate so I hope this explanation will help.

The goal of SS3 is to provide a recruitment time series that is unbiased regarding its arithmetic mean.  The deviations are relative to the spawner-recruitment curve which is scaled by the ln(R0) parameter.  Let's just think of them as devs from ln(R0) to make this easier.  The annual deviations in recruitment are lognormal so the mean of the resultant recruitments will be greater than exp(ln(R0)) unless a bias adjustment is applied.

• In years with no information about devs, including the forecast years, all the devs will be estimated to be near 0.0 and their geometric mean will be the same as their arithmetic mean, so no bias adjustment is necessary.  
• In years with high quality information about recdevs, the estimated devs will vary from year-to-year with a between year standard deviation that is near sigmaR, the true recruitment standard deviation.  In this case, the devs, which are in log space, need to be centered around a bias adjusted center such that the arithmetic mean of resultant recruitments will be correct.
• We never have perfect information about the recdevs, so we commonly apply a fraction of the theoretical maximum bias adjustment, commonly this is about 0.8.  r4ss estimates this based on the estimated residual variance of of the recdevs.

What we expect to see is a bias-adjustment time series that starts near 0.0 because applications with long time series have very little info about recdevs in early years, as data begin informing the recdevs, the bias adjustment increases.  Then near end of time series the recdevs again commonly go back to near zero so need no bias adjustment.  This continues into the forecast so there is no bias adjustment in the forecast either.

I am not certain exactly what you are referring to in your second sentence.  The set of controls for bias adjustment is in the control file and is shown below.

5 #_last_yr_nobias_adj_in_MPD; begin of ramp
 21 #_first_yr_fullbias_adj_in_MPD; begin of plateau
 92 #_last_yr_fullbias_adj_in_MPD
 98 #_end_yr_for_ramp_in_MPD (can be in forecast to shape ramp, but SS3 sets bias_adj to 0.0 for fcast yrs)
 0.80 #_max_bias_adj_in_MPD (typical ~0.8; -3 sets all years to 0.0; -2 sets all non-forecast yrs w/ estimated recdevs to 1.0; -1 sets biasadj=1.0 for all yrs w/ recdevs).

If your application needs a different mean recruitment in the forecast, you can use this control in the forecast.ss file:

0 # Forecast base recruitment:  0= spawn_recr; 1=mult*spawn_recr_fxn; 2=mult*VirginRecr; 3=deprecated; 4=mult*mean_over_yr_range
# for option 4, set phase for fore_recr_devs to -1 in control to get constant mean in MCMC, else devs will be applied.

I will try to respond again with a graphic showing the bias adjustment process.

Rick