merrillrudd/datamoderate

Stock assessment for catch and length data

datamoderate study

Folder overview

Simulation study located in sim_new folder with R code sim_SS.R.

Model runs in life history scenario folders

• Life history scenarios (e.g. short_slow)
• Fishing mortality scenario (e.g F1)
• Recruitment variability scenario (e.g. LowSigmaR)
• Files used to generate data in folder files
• Simulation replicate, or iteration (e.g. 1)
• Output files for generated population -- true values found in om
• True population values in om/Report.sso
• Length composition in om/data.ss_new
• Sample individuals per year to generate length composition
• perfect.ss used SS_splitdat to find perfect information on the length composition
• ss3.dat sampled N number of individuals from the perfect length composition using multinomial
• Number of years of length data for estimation models -- all use ss3.dat but will adjust the number of years of length to include (e.g. L100 uses 100 years, L75 uses 75 years, L1 uses last year only)
• Recruitment estimation scenarios, with all files for model run within.

Folder R

Includes some helper functions.

Some additional details on the methods

Operating model runs initial life history, F, and recruitment scenario files without hessian to generate true population.

• Four life history scenarios
• short_slow = shorter-lived (max age = 30 years), slower-to-Linf (expected to live at asymptotic length for final 10% of life)
• short_fast = shorter-lived, faster-to-Linf (expected to live at asymptotic length for final 50% of life)
• long_slow = longer-lived (max age = 60 years), slower-to-Linf
• long_fast = longer-lived, faster-to-Linf
• Each with variable M and k but sharing linf = 55 cm, length at 50% selectivity = 36.3 cm, h = 0.7, t0 = -1
• One fishing mortality time series, representative of U.S. West coast nearshore stocks.
• Two recruitment scenarios: LowSigmaR sigmaR = 0.4 and HighSigmaR sigmaR = 0.8 (also explored deterministic)
• 100 simulation replicates of each life history, F, and recruitment scenario.
• Used to create the "true population", with values in Report.sso and information on length structure in data.ss_new

Data generation -- multinomial to sample from length structure

• For each life history scenario and simulation replicate, generate length data from each year (e.g. perfect, N000, N50):
• perfect information with 1000 length samples
• 100 samples (more representative)
• Changing from 100 to 50 samples over the data series
• Then choose the number of years to include in the model:
• All 100 years of length data with perfect information
• Final 75 years, 20 years, 10 years, and 1 year, all subset from the same sampling procedure so that the length data is the same, only the number of years varies.

Estimation model

• Stock Synthesis (and will be tested with LIME)
• Recruitment estimation scenarios:
• Unadjusted - using the default values for bias ramp from ss3sim
• Bias adjusted - using SS_fitbiasramp to use estimated bias ramp parameters
• No estimation - do no estimate recruitment deviates.

Performance

• Bias (median relative error) and precision (median absolute relative error)
• Interval coverage (proportion of iterations where true value lies within 50% confidence intervals; nominal coverage would be equal to 50%. Scenarios greater than 50% tend to over-estimate uncertainty, while scenarios less than 50% tend to under-estimate uncertainty.)