SRModels {FLCore}R Documentation

Stock-Recruitment models

Description

A range of stock-recruitment (SR) models commonly used in fisheries science are provided in FLCore.

Each method is defined as a function returning a list with one or more elements as follows:

model
Formula for the model, using the slot names (rec and ssb) to refer to the usual inputs.
logl
Function to calculate the loglikelihood of the given model when estimated through MLE (See fmle).
initial
Function to provide initial values for all parameters to be minimization algorithms called by fmle or nls. If required, this function also have two attributes, lower and upper, that give lower and upper limits for the parameter values, respectively. This is used by some of the methods defined in optim, like "L-BFGS-B".
The model<- method for FLModel-class can then be called with value being a list thus described, the name of the function returning such a list, or the function itself. See the examples below.

Usage

  bevholt()
  bevholt.ar1()
  bevholt.c.a()
  bevholt.c.b()
        bevholt.d()
        bevholt.ndc()
        bevholt.sv()
        geomean()
        logl.ar1(rho, sigma2, obs, hat)
        ricker()
        ricker.ar1()
        ricker.c.a()
        ricker.c.b()
        ricker.d()
        ricker.sv()
        segreg()
        shepherd()
        shepherd.ar1()
        shepherd.d()
        shepherd.d.ar1()
        shepherd.ndc()
        shepherd.ndc.ar1()
        sv2ab(steepness, vbiomass, spr0, model)

Arguments

rho Autoregression
sigma2 Autoregression
obs Observed values
hat estimated values
steepness Steepness.
vbiomass Virgin biomass.
spr0 Spawners per recruit at F=0, see spr0.
model character vector with model name, either 'bevholt' or 'ricker'.

Details

Several functions to fit commonly used SR models are available. They all use maximum likelihood to estimate the parameters and except the autoregressive models, all use a normal log distribution for the residuals. Apart from the model equation parameters the variance of the log-residuals is also estimated.

bevholt:
Beverton & Holt stock-recruitment model fit:

R = alpha*S / (beta + S)

'a' is the maximum recruitment (asymptotically) and 'b' is the stock level needed to produce the half of maximum recruitment a/2. (a, b >0).

geomean:
Constant Recruitment model fit, equal to the historical geometric mean recruitment.

R = (R_1*R_2*...*R_n)^(1/n) = exp(mean(log(R_1) + ... + log(R_n)))

ricker:
Ricker stock-recruitment model fit:

R = a*S*exp(-b*S)

'a' is related to productivity and 'b' to density dependence. 'a' is the recruit per stock unit at small stock levels. (a, b > 0).

segreg:
Segmented regression stock-recruitment model fit:

R = ifelse(S <= b, a*S, a*b)

'a' is the slope of the recruitment for stock levels below 'b', and a*b is the mean recruitment for stock levels above 'b'. (a, b > 0).

shepherd:
Sheperd stock-recruitment model fit:

R = a * S/(1 + (S/b)^c)

Generalized Beverton & Holt and Ricker models with one covariate. The covariate can be used for example to account for an enviromental factor that influences the recruitment dynamics. In the equations 'c' is the shape parameter and 'X' is the covariate.

bevholt.c.a:
Beverton & Holt stock-recruitment model with one covariate in the numerator.

R = a*(1 - c*X)*S / (b + S)

bevholt.c.b:
Beverton & Holt stock-recruitment model with one covariate in thedenominator.

R = a*S / (b*(1 - c*X) + S)

ricker.c.a:
Ricker stock-recruitment model with one multiplicative covariate.

R = a*(1-c*X)*S*e^{-b*S}

ricker.c.b:
Ricker stock-recruitment model with one covariate in the exponent.

R = a*S*e^{-b*(1-c*X)*S}

Depensatory models for Beverton and Holt, Ricker and Shepherd models:

bevholt.d:
Depensatory Beverton & Holt stock-recruitment model

R = a*S^c / (b + S^c)

0< a, b, c < 'Inf'.

bevholt.ndc:
Depensatory Beverton & Holt stock-recruitment model

R = a*S^(1 + c) / (b + S^(1 + c))

0< a, b< 'Inf' and -0.5 < c < 0.5.

ricker.d:
Depensatory Ricker stock-recruitment model.

R = a*S^c*e^{-b*S}

shepherd.d:
Depensatory Shepherd stock-recruitment model.

R = a * S^2/(1 + (S /b)^c)

shepherd.ndc:
Depensatory Shepherd stock-recruitment model.

R = a * (S - d)/(1 + ((S - d) /b)^c)

Beverton and Holt and Ricker stock recruitment models parameterized for steepness and virgin biomass:

bevholt.sv:
Fits Beverton and Holt stock-recruitment model parameterised for steepness and virgin biomass.

a = vbiomass*4*steepness/(spr0*(5*steepness-1.0))

b = a*spr0*(1.0 /steepness - 1.0)/4.0

ricker.sv:
Fits a ricker stock-recruitment model parameterized for steepness and virgin biomass.

a = exp(b*vbiomass)/spr0

b = log(5*steepness) /(vbiomass*0.8)

Models fitted using autoregressive residuals of first order:

bevholt.ar1, ricker.ar1, shepherd.ar1, shepherd.d.ar1, shepherd.ndc.ar1:
Beverton and Holt, Ricker and Shepherd stock-recruitment models with autoregressive normal log residuals of first order. In the model fit the corresponding stock-recruitment model is combined with an autoregressive normal log likelihood of first order for the residuals. If R_t i the observed recruitment and Rest_t is the predicted recruitment, an autoregressive model of first order is fitted to the log-residuals, x_t = log(R_t/Rest_t).

x_t = rho*x_t-1 + e

Where 'e' follows a normal distribution with mean 0, e ~ N(0, sigma_ar^2).

Some additional useful functions:

Bevholt.SV, Ricker.SV:
Returns the predicted values using Beverton and Holt or Ricker stock recruitment model parameterized for steepness and virgin biomass. The function calculates the 'a' and 'b' parameters of the original parameterization for these models for a given steepness and virgin biomass values using 'sv2ab' function. This then uses the original parameterization to calculate the predicted recruiment.
sv2ab:
Calculates 'a' and 'b' parameters for Beverton and Holt stock recruitment models from the corresponding steepness, virgin biomass and spr0 parameters, (spr0 = spawning at recruit at F= 0.0parameters).

Beverton and Holt:

a = frac{4 cdot vbiomass cdot steepness}{spr0*(5*steepness-1)}

{a = vbiomass*4*steepness/(spr0*(5*steepness-1))}

b = frac{a cdot spr0 (frac{1}{steepness} - 1)}{4}

{b = a*spr0*(1/steepness - 1)/4}

Ricker:

a = exp(b*vbiomass)/spr0

b = frac{log(5 cdot steepness)}{0.8 cdot vbiomass}

{b = log(5*steepness)/(vbiomass*0.8)}

ab2sv:
Calculates steepness and virgin biomass parameters for Beverton and Holt or Ricker stock-recruitment relationship from the corresponding 'a', 'b' and spr0 parameters.

Beverton and Holt:

steepness = frac{a cdot spr0}{4 cdot b + a cdot spr0}

{steepness = a*spr0/(4*b+a*spr0)}

vbiomass = (spr0*a*(5*steepness - 1))/(4*steepness)

Ricker:

steepness = 0.2 e^{0.8 cdot b cdot vbiomass}

{steepness = 0.2*exp(b*vbiomass*0.8)}

vbiomass = log(spr0*a)/b

logl.ar1:
Log-Likelihood of autoregressive residuals time series of first order. If Y_t is the observed time series and Yest_t is the predicted time series using certain model, the loglikelihood of the residuals (X_t = Y_t - Yest_t) is calculated using the autoregressive model parameters sigma2 and rho.

Author(s)

The FLR Team

References

Beverton, R.J.H. and Holt, S.J. (1957) On the dynamics of exploited fish populations. MAFF Fish. Invest., Ser: II 19, 533.

Needle, C.L. Recruitment models: diagnosis and prognosis. Reviews in Fish Biology and Fisheries 11: 95-111, 2002.

Ricker, W.E. (1954) Stock and recruitment. J. Fish. Res. Bd Can. 11, 559-623.

Shepherd, J.G. (1982) A versatile new stock-recruitment relationship for fisheries and the construction of sustainable yield curves. J. Cons. Int. Explor. Mer 40, 67-75.

See Also

FLSR-class, FLModel-class

Examples

# inspect the output of one of the model functions
bevholt()
names(bevholt())
bevholt()$logl

# once an FLSR model is in the workspace ...
data(nsher)

# the three model-definition slots can be modified
# at once by calling 'model<-' with
# (1) a list
model(nsher) <- bevholt()

# (2) a function returning such a list
model(nsher) <- bevholt

# or (3) the name of such a function
model(nsher) <- 'bevholt'
[Package FLCore version 2.2 Index]