One draw of a deviation surface from the process the estimation model penalizes

The reverse of Get_PE_loglik. A shared level is drawn once and written wherever it appears, at the sd of the slot the penalty reads it at. Under iid every level is its own normal. Under the random walk each level steps from the year before. A first year the estimation model gives a diffuse start (rw_init_sigma) keeps the replicate's own value, since that start leaves the level to the data, as the recruitment and F walks do; under NA it starts at the walk's own sd, as the penalty does. The 3D GMRF and the separable AR1 draw each population, region and sex's whole surface over years and bins from the form's precision, conditional on the cells the map fixes at zero, which is the density the penalty evaluates at them.

Usage

draw_growth_pe_surface(
  PE_model,
  map,
  pe_pars,
  bins,
  fit_devs = NULL,
  n_cond = 0,
  rw_init_sigma = NA
)

Arguments

PE_model

Integer process error code, as Get_PE_loglik reads it.

map

Integer array [pop, region, year, bin, sex] of the levels, NA where a cell is fixed.

pe_pars

Array [pop, region, slot, sex], the half of growth_pe_pars the form reads.

bins

Bins the correlated forms run over.

fit_devs

The fit's deviations, shaped as map, read in the first n_cond years only.

n_cond

Integer. Years that keep the fit's deviations. Default 0 draws every year.

rw_init_sigma

The sd the estimation model gives a walk's first year, or NA (default) for the walk's own.

Value

Array shaped as map. A cell the map fixes keeps the fit's value, as the penalty reads it, zero when no fit is given.

Details

The first n_cond years keep the fit's deviations, and the years after them are drawn given those: a random walk steps on from the fit's last value, and the correlated forms draw from the precision conditional on the fit's cells as well as the fixed ones.