Two Approaches to Modelling

1
Two Approaches to Modelling

• Bruce McAdam
• The University of Aberdeen
• With Daniel Howell, Christian von Dorrien and
  Tara Marshall

2
UNCOVER Models

• We want to explore
• The effects of different management strategies on
  fish populations
• The effects of different aspects of
  biology/ecology/environment on fish populations
• To answer these questions we need
• Models with different degrees of realism
• Models implementing different management
  strategies
• We compare the outcomes of different models

3
For Example

• How will climate change affect chances of stock
  recovery?
• Model with two different climate scenarios
• Requires a model that depends realistically on
  climate
• Does improved understanding of SR relationship
  affect expectations of recovery?
• Model a simple SR relationship
• And a more complex one
• Compare projections.

4
Practical Matters

• We need to go from our understanding of the
  fisheries
• And create working models, in the form of
  computer programs.
• Generally we will use FLR to do this, but there
  are many design decisions to make

5
Two Approaches

• Start with a simple model of biology
• Make it more complex
• This could be done in R
• This lets us compare different models of ecology
• FLR also makes adding different management
  strategies easy
• Start with a (preexisting) complex model
• Integrate it into FLR in order to add different
  management strategies etc.

6
Examples

• You want to know whether using a more detailed SR
  relation matters
• start with a simple model in R (simple SR
  relation, e.g. current ICES assumptions)
• Make it more complex (a different SR relation,
  e.g. incorporating temperature)
• Compare results
• You want to compare two HCRs using the best
  possible biological information
• And you already have a detailed model of the
  stock
• Call the existing model from FLR
• Write R code for the different harvest rules and
  compare results.

7
Which to Use?

• It depends on
• What question you are asking
• What prior work has been performed
• What have you already got?
• What skills are available

8
An Example

• There are Gadget models of the Barents Sea
  ecosystem
• These run from a stand-alone application
• They are too complex to reimplement in R
• We would like to experiment with different
  environmental scenarios and management strategies
• So we need to use the Gadget model from within R

9
Gadget

• Its a model that can run simulations
• Given files describing the stocks, and giving
  parameters for the model
• It can also be used for model fitting
• Given description of model, and observations of
  the system, work out the parameters
• We are using it to run simulations only
• Previous work (other projects) have done the
  model fitting
• It runs as a stand-along application (.exe file)
  that reads and writes text files.

10
Youve seen this before
11
We use it like this
Gadget
FLR
12
Implementation

• Implement in R
• But, whenever we need a biological simulation
• Create Gadget input files
• Run the Gadget command
• Read data from Gadget output files
• This contains actual stock and catch data
• Gadget is treated as a black box
• R knows nothing about the sort of model it is.

13
Algorithm (written in R)

• Start with a folder containing all of Gadgets
  input files
• Run Gadget to generate output files
• Main Loop (repeat for some number of years)
• Read output files
• Create necessary FLR objects
• Run an assessment (e.g. XSA), apply a harvest
  control rule (HCR)
• Update Gadget input files with next years
  catch/environment (and anything else needed)
• Run Gadget with new input, and repeat the loop
• After finishing, plot graphs etc.

14
Preliminary Results

• The cod model is complex, so we are starting with
  a simpler (Icelandic) haddock model
• We model a constant recruitment scenario, with
  different levels of TAC (fixed percentage of
  SSB).
• This is all just experimentation to check it will
  be possible to implement useful models

15
Pretty Pictures
TAC10 of SSB
TAC50 of SSB
16
Technical Issues

• Weve done about 2 person-weeks of work
• We had to implement
• Reading data from Gadget
• Writing data to Gadget
• Running Gadget
• Converting Gadget data into an FLR form (FLStock
  objects)
• Running an assessment on the FLStocks
• Surprisingly, we found it much easier to
  communicate with Gadget, than to do the
  assessment in FLR (work in progress)

17
Where Next?

• Still to get assessments working for the haddock
  model
• FLXSA
• Then get it working with cod
• Shouldnt be too much work
• And look (a bit) at efficiency

18
Summary

• There are two different approaches to modelling
• Write a model in R, and build it up
• Start with a complex model, and call it from R
• FLR can call external models
• Its not too difficult
• You need to do quite a lot of programming in R
  but not as much as implementing the model
• The code you write will be case specific
• The model doesnt have to be Gadget.

19
Remember

• My job is to help people customise models like
  this
• But you need to keep me informed about your
  objectives etc.
• And make sure you use me as a resource
• Otherwise, the project will not make efficient
  use of my time (I work 100 on UNCOVER).