FPRESS

… need some nice text here

FPRESS was conceived by Cirian Kelly and Andrew Campbell at Marine Institute, Galway. The generic source code (version FPRESS_2.0-0) was obtained from Andrew Campell, 12.8.2009. Permission to make the this version available on this DokuWiki is pending.

It should be noted that the FPRESS_2.0-0 code is “rather old” and has not been maintained by the original team. Or as stated by Andrew Campell in e-mail correspondence: ”This version of FPRESS is rather old at this stage. Despite my best intentions, it has been very difficult to find the means and the time to maintain a generic version of the model for distribution. Work has tended to focus on developing it for specific applications such as the NEA Mackerel LTM (as you mentioned below) and the Western Horse Mackerel Management Plan development. It has also been used for Irish Sea Cod and Sardine work.” This means that numerous “bugs” that may have been discovered in later works have not been implemented/corrected in the generic code. Nor for that matter some improvements/additions that have been made. Given this the intent of this DokuWiki space is to provide a forum to provide some communication with regards to updates/developments with regards to the FPRESS code. As a start it is suggested that the space is structured with the following sections (some will need creation of new pages within this namespace - some specifics are provided here):

1. The absolute minimum changes needed to correct some bugs in FPRESS 2.0.0.
2. Suggestions for restructuring of the code with the objective of making it more modular.
3. Recent additional implementations
4. Case specific adaptation

The following is a list of changes made to the generic code.

run.r The following changes are absolutely necessary (has been confirmed by Andrew), otherwise the script will halt:

• line 427: replaced ncol=18 with ncol=17
• line 954: replaced newfbias with newvarbias
• …

Additional error codes not yet in the changed script above (will do once confirmed with author):

• line 874: replaced fmult with 1
• …

Changes in code related to information provided to users:

• line 244: replaced ssbhcr with fbarhcr
• line 249: replaced ssbhcr with minimumtac

Changes in code because of redundancy (i.e. not really necessary):

• line 46: commented out, is redundant (see, line 299 - where the whole process to accumulate savedobjs is restarted by setting savedobjs←c())
• line 254: commented out, is redundant (see comment on line 45, here above)
• line 258: commented out (same as above)
• line 262: commented out (same as above)
• line 270: commented out (same as above)
• line 274: commented out (same as above)
• line 278: commented out (same as above)
• line 282: commented out (same as above)
• line 286: commented out (same as above)
• line 290: commented out (same as above)
• line 294: commented out (same as above)

The following changes are needed if stochastic hockey-stick is used as a recruitment model

• line 416: replace reccvd with reccv
• …

A question: It seems that the stochastic recruitment in FPRESS 2.0.0 is normally distributed. A case in point is e.g. Recruitment function 5 [function recruit5() in recruits.r] There one has a transformed parameter fit (implicitly implying log-normal distribution of recruitment), but the stochastic recruitment in the simulation is normally distributed, implemented on nontransformed R-values. This thought/question with regards to the FPRESS implementation is supported in the algorithm of in the summaryplot.r where normal distribution is assumed when plotting the 95% and 99% ci.

This default in the FPRESS code is of concern. Recruitment by default is log-normally distributed (see Myers meta database, Hilborn and Walters 1994, Haddon 2002). And the parameter estimates within FPRESS Ricker function assume that it is so. In case specific examples, one may argue that a normal distribution may be justified, but the default in forward simulation should be log-normal.

A simple case to illustrate FPRESS default is the following R-code:

n <- 10000
avg <- 200
cv <- 0.6

# FPRESS approoach
x <- rnorm(n,avg,cv*avg)
hist(x)

# The conventional default
x <- avg*rlnorm(n=1000,0,cv)
hist(x)

Makes a difference when it comes to risk evaluation, which algorithm is applied for future stochastic R scenarios. Please correct this if it is wrong.

An bundled source folder taking into the account the changes made above can be found here.

• Note that the changed run and recruit files now have different names, for other scripts the original file names have been kept. To be more specific:
  • the original run.r has been replaced with run201.r
  • the original recruit.r has been replaced with recruit201.r

In generic FPRESS 2.0.x code provides ample commentaries in individual places, put in place for outsiders to obtain a better understanding of the code.

However, ..

… to be filled in.

… to be filled in.