Skip to contents

Seasonal sampling design

Source: R/seasons.R
season_design.Rd

Describes how the annual length and biological samples of a fleet are spread over seasons. The default, four seasons with equal shares, is a balanced design. Imbalanced designs:

Usage

season_design(
  n = 4,
  share = NULL,
  shift = NULL,
  kappa = Inf,
  n_mult = NULL,
  p_missing = 0,
  labels = NULL
)

Arguments

n

number of seasons (1 = one annual sample at mid-year, as in FLOSS <= 0.4).

share

sample share by season (rescaled to sum to 1); default equal.

shift

optional list of from (year), share and optionally over (years): the shares change from from on, linearly over over years (default 0, a step). Several shifts as a list of such lists.

kappa

Dirichlet concentration for random year-to-year shares; Inf (default) = no random variation.

n_mult

optional list of from, mult and optionally over: the annual sample size (clusters, fish measured, biological samples) is multiplied by mult from from on (ramped over over years).

p_missing

probability that a year is not sampled at all.

labels

season names; default "S1", "S2", ...

Value

a list of class floss_seasons.

Details

  • fixed imbalance: unequal share, e.g. c(0.5, 0.3, 0.2, 0) (most samples early in the year, none in season 4, e.g. a monsoon closure of the landing site);

  • shift: the shares change from a given year (shift), e.g. a new observer programme that samples in the second half of the year only;

  • opportunistic: the shares vary at random from year to year around share (kappa, the concentration of a Dirichlet distribution: 50 = small departures, 5 = large ones), as when samples are taken whenever an observer happens to be available.

Sampling intensity over the years:

  • intensity change: the annual sample size is multiplied by n_mult from a year on (list(from = 2020, mult = 0.3): a funding cut), with an optional ramp over over years;

  • missing years: each year is not sampled at all with probability p_missing.

Within a season, the age composition of the catch is the Baranov catch between the start and end of the season (fishing effort and natural mortality constant through the year), and lengths are those at the middle of the season (von Bertalanffy growth with age + season mid-point). Because effort is constant within the year, the catch of each season is known from the dynamics, and equal season weights (lfd_annual()) are the correct way to pool unevenly spread samples.

Examples

season_design()                                   # balanced, 4 seasons
#> FLOSS seasonal sampling: 4 season(s) (balanced) 
#>   share: 0.25 0.25 0.25 0.25 
season_design(share = c(0.6, 0.3, 0.1, 0))         # fixed imbalance
#> FLOSS seasonal sampling: 4 season(s) (imbalanced) 
#>   share: 0.60 0.30 0.10 0.00 
season_design(shift = list(from = 2015, share = c(0, 0, 0.5, 0.5)))
#> FLOSS seasonal sampling: 4 season(s) (imbalanced) 
#>   share: 0.25 0.25 0.25 0.25 
#>   from 2015 : 0.00 0.00 0.50 0.50 
season_design(kappa = 5)                           # opportunistic
#> FLOSS seasonal sampling: 4 season(s) (imbalanced) 
#>   share: 0.25 0.25 0.25 0.25 
#>   random year-to-year shares, kappa = 5