fcs2RunINLA: Fit Approximate FCS2 Models with INLA

In aquaMetrics/fcs2: Fisheries Classification Scheme 2 For SNIFFER

Description

Fits two approximate FCS2 sub-models using Integrated Nested Laplace Approximations (INLA).

Usage

fcs2RunINLA(fit, run = TRUE, verbose = TRUE)

Arguments

fit	an object of class "fcs2Fit", usually returned by fcs2FitModel.
run	which approximate FCS2 models to fit with INLA. If TRUE (the default) or "both", INLA is run for both models. Alternatively, if "rho" or "prevalence" the prevalence model only is run and if "mu" or "abundance" the approximate abundance model only is run. The abundance model can only be run if the prevalence model is either run first or is provided through fit.
verbose	whether to print progress to screen.

Details

INLA is used to fit two models that together approximate the full FCS2 statistical model (see fcs2FitModel.

Firstly, an approximate presence/absence model is fitted to estimate the prevalence regression. This assumes the observed presence follows a Bernoulli distribution with probability of presence ρ (the prevalence). The logit regression for ρ takes the same form as in the full FCS2 model.

After fitting the prevalence sub-model, the presence of the species can be estimated. Data for surveys either observed or predicted to be present are then modelled by the second model that attempts to estimate the abundance regression. This models the total catch T over all passes by the Negative Binomial distribution with shape r and mean given by a μ, where a is the survey area and μ is the abundance. The regression for μ takes the same form as in the full FCS2 model.

This model is exact for the original FCS2 model but only an approximation of the total catch when present in the multiple-pass model since it ignores the dependence upon the catch probability q and the number of runs d. However, since q is constant and if the number of runs does not vary significantly, this discrepancy should be mostly absorbed into an error in the constant parameter β_0. If this is true, the model estimates can still be used to estimate the regression terms so that their significance can be judged.

Value

a list with two components:

rhoFit	an "inla" object containing the approximate prevalence model fit
muFit	an "inla" object containing the approximate abundance model fit, or NULL if this was not calculated or provided

See inla from the package INLA for a description of the "inla" object.

Note

The user would not usually call this function directly as it is called if requested by fcs2FitModel.

See Also

fcs2FitModel for the preferred way of calling this function.
inla from the package INLA for details of the INLA estimation method and the "inla" objects it returns.

aquaMetrics/fcs2 documentation built on Aug. 21, 2021, 12:55 p.m.