Running the MSE

Once an operating model (OM) object, a recruitment object, and a management procedure (MP) object have all been appropriately defined, the MSE simulation loop can be almost be initiated.

MSE Options

A list of MSE options must be defined to control internal aspects of the MSE simulation loop, primarily how long to run the MSE for. The MSE options object requires the following inputs:

• n_proj_years - the number of projection years to run the MSE for. This is the number of years for which the specified management procedure will be applied.
• n_spinup_years - the number of years for which to apply the specified OM before, applying the specified MP. For sablefish, this parameter is the number of years since 1960 (0-64), and indicates the calendar year you would like to begin MSE simulations from.
• recruitment_start_year - which year of the simulation new recruitment events should begin in. Often this is the same as n_spinup_years but can be longer if the user wants to use a specific recruitment timeseries at the beginning of the MSE simulations.
• run_estimation - whether the estimation method should run or not. If FALSE the management procedure will be applied to OM outputs directly rather than to estimates from the EM.

Sensible defaults are available via the setup_mse_options function. Using them will begin the MSE simulations from the estimated population state in 2024, and project forward for 100 years.

options <- setup_mse_options()
options

## $n_proj_years
## [1] 100
## 
## $n_spinup_years
## [1] 64
## 
## $recruitment_start_year
## [1] 64
## 
## $run_estimation
## [1] TRUE

Note that the total number of simulation years for a single MSE run is n_spinup_years + n_proj_years.

Parallel MSE Simulations

Frequently, it is desirable to run multiple MSE simulations across different random seeds, to evaluate an MP in the face of uncertainty.

The run_mse_parallel(...) function is designed to act as a simple parallel wrapper for the base run_mse(...) function that will run multiple MSE simulations, with the same OM and MP specifications, where each simulation will only vary by the random seed used to generate annual recruitment levels and simulate observations.

To use run_mse_parallel users must specify the number of simulations to run, and provide a list of simulation seeds in addition to the OM, MP, and options objects that are required for the base run_mse function. run_mse_parallel will handle parallelizing the MSE simulations and aggregating the output data as simulations complete.

run_mse_parallel(
    nsims = 10,
    seeds = sample(1:1000, 10),
    om=om,
    hcr=mp,
    mse_options=options,
    nyears = 1
)

Note that this function will use up to N-2 available compute cores on a machine, so as to allow the machine to remain functional. Parallel processing is enabled through use of the pblapply package and associated function. Users DO NOT need to handle creating their own parallel processing cluster; this is handled internally.

Multiple MSE Simulations

Often, multiple combinations of MPs and OMs are tested together within an MSE to evaluate MPs across different states of nature. A wrapper function run_mse_multiple is provided to facilitate this and return appropiately formatted MSE outputs.

To use run_mse_multiple, users should define multiple OM, MP, and/or MSE options objects, place them into lists, and pass the respective lists to run_mse_multiple:

run_mse_multiple(
    # List of OMs to run
    om_list = list(om1, om2),
    # List of MPs to run
    hcr_list = list(mp1, mp2),
    # List of MSE options to run
    mse_options_list = list(options)
    seed_list = sample(1:1000, 10),
    nyears=1
)

run_mse_multiple will run all combinations of OMs, MPs, and MSE Options provided in their respective lists, each for the number of simulations provided in seed_list.