Intended for conditioning operating models for MSEtool. For data-limited stocks, this function can generate a range of potential depletion scenarios inferred from sparse data.
From a historical time series of total catch or effort, and potentially age/length compositions and multiple indices of abundance, the RCM returns a range of values for depletion, selectivity,
unfished recruitment (R0), historical fishing effort, and recruitment deviations for the operating model. This is done by sampling life history parameters
provided by the user and fitting a statistical catch-at-age model (with the predicted catch equal to the observed catch).
Alternatively one can do a single model fit and sample the covariance matrix to generate an operating model with uncertainty based on the model fit.
Either a full catch (conditioned on catch) or effort (conditioned on effort) time series is needed but missing data (as NAs) are allowed for all other data types.
check_RCMdata evaluates whether the inputs in the S4 RCMdata object are correctly formatted.
check_RCMdata(RCMdata, OM, condition = "catch", silent = FALSE)RCM(OM, data, ...)
# S4 method for OM,RCMdata
RCM(
OM,
data,
condition = "catch",
selectivity = "logistic",
s_selectivity = NULL,
LWT = list(),
comp_like = c("multinomial", "lognormal", "mvlogistic", "dirmult1", "dirmult2"),
prior = list(),
max_F = 3,
cores = 1L,
integrate = FALSE,
mean_fit = FALSE,
drop_nonconv = FALSE,
drop_highF = FALSE,
control = list(iter.max = 2e+05, eval.max = 4e+05),
start = list(),
map = list(),
silent = FALSE,
...
)
# S4 method for OM,list
RCM(
OM,
data,
condition = "catch",
selectivity = "logistic",
s_selectivity = NULL,
LWT = list(),
comp_like = c("multinomial", "lognormal", "mvlogistic", "dirmult1", "dirmult2"),
ESS = c(30, 30),
prior = list(),
max_F = 3,
cores = 1L,
integrate = FALSE,
mean_fit = FALSE,
drop_nonconv = FALSE,
drop_highF = FALSE,
control = list(iter.max = 2e+05, eval.max = 4e+05),
start = list(),
map = list(),
silent = FALSE,
...
)
# S4 method for OM,Data
RCM(
OM,
data,
condition = "catch",
selectivity = "logistic",
s_selectivity = NULL,
LWT = list(),
comp_like = c("multinomial", "lognormal", "mvlogistic", "dirmult1", "dirmult2"),
ESS = c(30, 30),
prior = list(),
max_F = 3,
cores = 1L,
integrate = FALSE,
mean_fit = FALSE,
drop_nonconv = FALSE,
drop_highF = FALSE,
control = list(iter.max = 2e+05, eval.max = 4e+05),
start = list(),
map = list(),
silent = FALSE,
...
)
An object of class RCModel (see link for description of output).
check_RCMdata returns a list of updated RCMdata object, OM, and StockPars and FleetPars from the Hist object generated
from the OM.
An RCMdata object.
An object of class MSEtool::OM that specifies natural mortality (M), growth (Linf, K, t0, a, b), stock-recruitment relationship, steepness, maturity parameters (L50 and L50_95), standard deviation of recruitment variability (Perr), as well as index uncertainty (Iobs).
String to indicate whether the RCM is conditioned on "catch" (where F are estimated parameters), "catch2" (where F is solved internally using Newton's method),
or "effort" (F is proportional to an index series in data@Ehist. Can be fleet specific, in which case some combination of "catch" and "effort" are permissible.
Logical to indicate whether informative messages will be reported to console.
Data inputs formatted in a RCMdata (preferred) or MSEtool::Data object. Use of a list is deprecated. See Data section below.
Other arguments to pass in for starting values of parameters and fixing parameters. See details.
A character vector of length nfleet to indicate "logistic_length", "dome_length", "logistic_age", "dome_age", or "free" selectivity for each fleet in Chist.
If there is time-varying selectivity, this is a character vector of length nsel_block (see Data section below). "free" indicates independent selectivity parameters for each age,
and additional modifications for fixing selectivity parameters will likely be needed. See Additional arguments section.
A vector of length nsurvey to indicate the selectivity of the corresponding columns in data$Index. Use "B" for
total biomass, or "SSB" for spawning biomass (by default, "B" is used). Use numbers if the survey selectivity follows a fleet (corresponding to the columns in data$Chist, e.g., 1 = first fleet/column and so on).
If the survey selectivity is otherwise independent of anything else in the model, use "logistic_length", "dome_length", "logistic_age", "dome_age", or "free" to specify the functional form of selectivity, and
see Additional arguments section for setup of survey selectivity parameters and Articles section for more information.
A named list of likelihood weights for the RCM. See below.
A string indicating the statistical distribution for the composition data, either "multinomial" (default), "lognormal", "mvlogistic" (multivariate logistic),
"dirmult1" (Dirichlet multinomial, linear version), or "dirmult2" (saturating version; see Thorson et al. 2017).
A named list for the parameters of any priors to be added to the model. See below.
The maximum F for any fleet in the scoping model (higher F's in the model are penalized in the objective function). This argument will also update OM@maxF. See also drop_highF.
Integer for the number of CPU cores (set greater than 1 for parallel processing).
Logical, whether to treat recruitment deviations as penalized parameters in the likelihood (FALSE) or random effects to be marginalized out of the likelihood (TRUE).
Logical, whether to run an additional with mean values of life history parameters from the OM.
Logical, whether to drop non-converged fits of the RCM, including fits where F = NA.
Logical, whether to drop fits of the RCM where F = max_F.
A named list of arguments (e.g, max. iterations, etc.) for optimization, to be passed to the control argument of stats::nlminb().
A list of starting values for the TMB model. See details.
A list of map argument to TMB models to override defaults. See MakeADFun and details.
A vector of length two. A shortcut method to setting the maximum multinomial sample size of the age and length compositions. Not used when data are provided in a RCMdata object.
The following priors can be added as a named list, e.g., prior = list(M = c(0.25, 0.15), h = c(0.7, 0.1).
For each parameter below, provide a vector of values as described:
R0A vector of length 3. The first value indicates the distribution of the prior: 1 for lognormal, 2 for uniform
on log(R0), 3 for uniform on R0. If lognormal, the second and third values are the prior mean (in normal space) and SD (in log space).
Otherwise, the second and third values are the lower and upper bounds of the uniform distribution (values in normal space).
hA vector of length 2 for the prior mean and SD, both in normal space. Beverton-Holt steepness uses a beta distribution, while Ricker steepness uses a normal distribution.
MA vector of length 2 for the prior mean (in normal space) and SD (in log space). Lognormal prior.
qA matrix for nsurvey rows and 2 columns. The first column is the prior mean (in normal space) and the second column
for the SD (in log space). Use NA in rows corresponding to indices without priors.
See online documentation for more details.
Several articles are available for RCM:
Setup of selectivity settings and index catchability (useful for more data-rich cases)
One of indices, age compositions, or length compositions should be provided in addition to the historical catch or effort. Not all arguments are needed to run the model (some have defaults, while others are ignored if not applicable depending on the data provided).
The data variable can be an object of class RCMdata. See help file for description of inputs.
Alternatively, the data input can be a MSEtool::Data S4 object which will retrieve data from the following slots:
Data@Catcatch series (single fleet with the Data S4 object)
Data@Efforteffort series
Data@CAAfishery age composition
Data@CAL, Data@CAL_midsfishery length composition and corresponding length bins
Data@Ind, Data@SpInd, Data@VInd, Data@AddIndindices of abundance
Data@CV_Ind, Data@CV_SpInd, Data@CV_VInd, Data@CV_AddIndannual coefficients of variation for the corresponding indices of abundance. CVs will be converted to lognormal standard deviations.
Data@MLfishery mean lengths
Data@AddIndV, Data@AddIndType, Data@AddIunitsAdditional information for indices in Data@AddInd:
selectivity and units (i.e., biomass or abundance).
There is no slot in the Data S4 object for the equilibrium catch/effort. These can be passed directly in the function call, i.e., RCM(OM, Data, C_eq = C_eq, ...).
Use of a list is deprecated. For backwards compatibility, here is the list of supported entries:
ChistA vector of historical catch, should be of length OM@nyears. If there are multiple fleets: a matrix of OM@nyears rows and nfleet columns.
Ideally, the first year of the catch series represents unfished conditions (see also C_eq).
C_sdA vector or matrix of standard deviations (lognormal distribution) for the catches in Chist.
If not provided, the default is 0.01. Only used if condition = "catch".
EhistA vector of historical effort, should be of length OM@nyears (see also E_eq).
IndexA vector of values of an index (of length OM@nyears). If there are multiple indices: a matrix of historical indices of abundances, with rows
indexing years and columns indexing the index.
I_sdA vector or matrix of standard deviations (lognormal distribution) for the indices corresponding to the entries in Index.
If not provided, this function will use values from OM@Iobs.
I_typeObsolete as of version 2.0. See s_selectivity argument.
CAAFishery age composition matrix with nyears rows and OM@maxage+1 columns. If multiple fleets: an array with dimension:
nyears, OM@maxage, and nfleet.
CALFishery length composition matrix with nyears rows and columns indexing the length bin. If multiple fleets:
an array with dimension: nyears, length bins, and nfleet.
MSA vector of fishery mean size (MS, either mean length or mean weight) observations (length OM@nyears),
or if multiple fleets: matrix of dimension: nyears, nfleet.
Generally, mean lengths should not be used if CAL is also provided, unless mean length and length comps are independently sampled.
MS_typeA character (either "length" (default) or "weight") to denote the type of mean size data.
MS_cvThe coefficient of variation of the observed mean size. If there are multiple fleets, a vector of length nfleet.
Default is 0.2.
s_CAASurvey age composition data, an array of dimension nyears, maxage+1, nsurvey.
s_CALSurvey length composition data, an array of dimension nyears, length(length_bin), nsurvey.
length_binA vector for the midpoints of the length bins for CAL and s_CAL. All bin widths should be equal in size.
C_eqA numeric vector of length nfleet for the equilibrium catch for each fleet in Chist prior to the first year of the operating model.
Zero (default) implies unfished conditions in year one. Otherwise, this is used to estimate depletion in the first year of the data. Alternatively,
if one has a full CAA matrix, one could instead estimate "artificial" rec devs to generate the initial numbers-at-age (and hence initial
depletion) in the first year of the model (see additional arguments).
C_eq_sdA vector of standard deviations (lognormal distribution) for the equilibrium catches in C_eq.
If not provided, the default is 0.01. Only used if condition = "catch".
E_eqThe equilibrium effort for each fleet in Ehist prior to the first year of the operating model.
Zero (default) implies unfished conditions in year one. Otherwise, this is used to estimate depletion in the first year of the data.
abs_IOptional, an integer vector to indicate which indices are in absolute magnitude. Use 1 to set q = 1,
otherwise use 0 to estimate q.
I_unitsOptional, an integer vector to indicate whether indices are biomass based (1) or abundance-based (0). By default, all are biomass-based.
age_errorOptional, a square matrix of maxage + 1 rows and columns to specify ageing error. The aa-th column assigns a proportion of the true age in the a-th row to observed age. Thus, all rows should sum to 1. Default is an identity matrix (no ageing error).
sel_blockOptional, for time-varying fleet selectivity (in time blocks), a integer matrix of nyears rows and nfleet columns
to assigns a selectivity function to a fleet for certain years.
For RCM, additional arguments can be passed to the model via ...:
plusgroupLogical for whether the maximum age is a plusgroup or not. By default, TRUE.
fix_domeLogical for whether the dome selectivity parameter for fleets is fixed. Used primarily for backwards compatibility,
this is overridden by the map argument.
resampleLogical, whether the OM conditioning parameters (recruitment, fishing mortality, SSB, selectivity, etc.) are obtained by sampling the Hessian matrix from a single model fit. By default FALSE. This feature requires identical biological parameters among simulations.
pbc_recdevVector of length nyears. Proportion of the bias correction to apply annually to the recruitment deviations (if estimated).
The bias correction from logspace to normal space is exp(log_rec_dev[y] - 0.5 * pbc_recdev[y] * sigmaR^2). Default proportion is 1.
Starting values can be specified in a named list for the following:
vul_parA matrix of 3 rows and nfleet columns for starting values for fleet selectivity. The three rows correspond
to LFS (length of full selectivity), L5 (length of 5 percent selectivity), and Vmaxlen (selectivity at length Linf). By default,
the starting values are values from the OM object. If any selectivity = "free", then this matrix needs to be of maxage+1 rows where
the row specifies the selectivity at age. See Articles section.
ivul_parA matrix of 3 rows and nsurvey columns for starting values for fleet selectivity. Same setup as vul_par. Values in the column are ignored
if s_selectivity is mapped to a fishing fleet (add NA placeholders in that case).
If any s_selectivity = "free", then this matrix needs to be of maxage+1 rows where
the row specifies the selectivity at age.
log_rec_devA numeric vector of length nyears for the starting values of the log-recruitment deviations.
log_early_rec_devA numeric vector of length OM@maxage for the starting values of the recruitment deviations controlling the abundance-at-age in the first year of the model.
qA numeric vector of length nsurvey for index catchability. See online article for more information.
Parameters can be fixed with the map argument (also a named list, corresponding to the start list). Each
vector or matrix in the map argument will be the same dimension as in the start entry. If an entry is NA, the corresponding parameter is fixed in the model to the starting
value. Otherwise, an integer for each independent parameter, i.e., shared or mirrored parameters get the same integer entry.
vul_parAn integer matrix of the same dimension as start$vul_par. By default, selectivity is fixed if there are no age or length composition for that fleet
or survey, otherwise estimated. Unused cells in the start$vul_par matrix should be given NA in the map matrix.
ivul_parThe map argument for the survey selectivity parameters (same dimension as start$ivul_par). Placeholder parameters should have a map value of NA.
log_early_rec_devA vector of length OM@maxage that indexes which recruitment deviates for the cohorts in the first year of the model are fixed (using NA) or estimated (a separate integer).
By default, no deviates are estimated (all are NA).
log_rec_devA vector of length OM@nyears that indexes which recruitment deviates are fixed (using NA) or estimated (a separate integer).
By default, all these deviates are estimated.
qA vector of length nsurvey for index catchability. q should be an estimated parameter when sharing across surveys (perhaps with differing selectivity). Otherwise, it is solved analytically
where individual parameters are independent of other indices. Use RCMdata@abs_I for fixing the catchability to 1. See online article for more information.
LWT is an optional named list containing the likelihood weights (values >= 0) with the possible options:
Chist, CAA, CAL, MS, C_eq: A vector of length nfleet for each.
Index, IAA, IAL: A vector of length nsurvey for each.
By default, all likelihood weights are equal to one if not specified by the user.
Annual multinomial sample sizes for the age and length comps can now be provided directly in the
RCMdata object. For a list or MSEtool::Data object, use the ESS argument.
Q. Huynh
Fleet selectivity is fixed to values sampled from OM if no age or length compositions are provided.
Survey selectivity is estimable only if IAA or IAL is provided. Otherwise, the selectivity should
be mirrored to a fleet (vulnerable biomass selectivity) or indexed to total or spawning biomass (see s_selectivity).
Parameters that were used in the fitting model are placed in the RCM@OM@cpars list.
If the operating model OM uses time-varying growth or M, then those trends will be used in the RCM as well.
Non-stationary productivity creates ambiguity in the calculation and interpretation of depletion and MSY reference points.
The easiest way to turn off time-varying growth/M is by setting: OM@Msd <- OM@Linfsd <- OM@Ksd <- c(0, 0).
To play with alternative fits by excluding indices, for example, or other optional data, set the corresponding likelihood weight to zero. The model will still generate the inferred index but the data won't enter the likelihood. See section on likelihood weights.
Thorson et al. 2017. Model-based estimates of effective sample size in stock assessment models using the Dirichlet-multinomial distribution. Fish. Res. 192:84-93. tools:::Rd_expr_doi("10.1016/j.fishres.2016.06.005")
plot.RCModel RCModel compare_RCM pcod RCM2MOM posterior
# \donttest{
# An example that conditions a Pacific cod operating model. There are 48 simulations,
# where values of natural mortality and steepness are sampled from distributions.
# The model is fitted with priors on the index catchability. Maturity and selectivity
# are knife-edge at the age of 2 years. See online tutorial for more information.
data(pcod)
mat_ogive <- pcod$OM@cpars$Mat_age[1, , 1]
out <- RCM(OM = pcod$OM, data = pcod$data,
condition = "catch", mean_fit = TRUE,
selectivity = "free", s_selectivity = rep("SSB", ncol(pcod$data@Index)),
start = list(vul_par = matrix(mat_ogive, length(mat_ogive), 1)),
map = list(vul_par = matrix(NA, length(mat_ogive), 1),
log_early_rec_dev = rep(1, pcod$OM@maxage)),
prior = pcod$prior)
plot(out, s_name = colnames(pcod$data@Index))
# Alternative OM with age-3 maturity and selectivity instead.
out_age3 <- local({
pcod$OM@cpars$Mat_age[, 2, ] <- 0
mat_ogive_age3 <- pcod$OM@cpars$Mat_age[1, , 1]
RCM(OM = pcod$OM, data = pcod$data,
condition = "catch", mean_fit = TRUE,
selectivity = "free", s_selectivity = rep("SSB", ncol(pcod$data@Index)),
start = list(vul_par = matrix(mat_ogive_age3, length(mat_ogive_age3), 1)),
map = list(vul_par = matrix(NA, length(mat_ogive_age3), 1),
log_early_rec_dev = rep(1, pcod$OM@maxage)),
prior = pcod$prior)
})
compare_RCM(out, out_age3, scenario = list(names = c("Age-2 maturity", "Age-3 maturity")),
s_name = colnames(pcod$data@Index))
Hist <- runMSE(out@OM, Hist = TRUE)
# }
Run the code above in your browser using DataLab