Skip to contents

impute_count

Source: R/rbms_index_modelling.r
impute_count.Rd

impute_count

Usage

impute_count(
  ts_season_count,
  ts_flight_curve,
  TimeUnit = "d",
  MultiVisit = "mean",
  weekday = 3,
  sp = NULL,
  YearLimit = NULL,
  SelectYear = NULL,
  CompltSeason = TRUE,
  SindexMethod = c("ratio", "glm"),
  GlmFit = c("newton", "ipf"),
  verbose = TRUE
)

Arguments

ts_season_count

data.table Time-series of counts for a specific species across all sites as returned by ts_monit_count_site.

ts_flight_curve

pheno_curve class object or data.table with light curves and relative abundances (NM) for a specific species as returned by flight_curve.

TimeUnit

character The time-step for which the spline should be computed, 'd' day or 'w' week.

MultiVisit

string Function to apply for summarising multiple counts within a time unit, 'max' or 'mean' (default).

weekday

Integer for selected day of the week for weekly summary, default is 3 (Wednesday). [1-7] where 1 = Monday.

sp

integer or string Species ID or name.

YearLimit

integer Define the span of years (+/- number of year) to look for a flight curve, if NULL no restriction is set.

SelectYear

integer Select a specific year to compute the flight curve, default=NULL.

CompltSeason

logical Restrict computation of flight curve for years where the complete season was sampled, default=TRUE.

SindexMethod

character Method to compute the site index, 'ratio' (default) or 'glm', see details.

GlmFit

character Fitting of the GLM when SindexMethod = 'glm', 'newton' (default) or 'ipf', see details.

verbose

a logical indicating if some “progress report” should be given.

Value

A data.table based on the entry count data, augmented with site indices 'SINDEX' and imputed weekly count 'IMPUTED_COUNT'. SINDEX is NA when the visits cover no part of the flight curve (TOTAL_NM = 0, e.g. no flight curve available within YearLimit).

Details

Site indices can be extracted from the data.table returned by this function. With SindexMethod = 'ratio', the site index is computed by adjusting the count by the proportion of the flight curve covered by the visits (TOTAL_COUNT / TOTAL_NM), each site-year being estimated on its own. With SindexMethod = 'glm', the counts of the time-steps not visited are predicted from a Poisson GLM with site and year effects and the flight curve as offset, COUNT ~ factor(SITE_ID) + factor(M_YEAR) + offset(log(NM)), and the site index is the sum of the observed and predicted counts. The GLM borrows the level of the site and the year from the other years and sites, which avoids extreme site indices when a small part of the flight curve is covered, but assumes no site x year interaction for the time-steps not visited. Sites where the species was never observed keep the ratio (0). The GLM is fitted without model matrix, on the counts summed per site-year. With GlmFit = 'newton', the site effects are concentrated out of the likelihood and Newton steps are made on the year effects only (few iterations); with GlmFit = 'ipf', iterative proportional fitting is used (more iterations when years are weakly linked). Both give the estimates of glm(), 'newton' uses 'ipf' if it fails. With 'newton', the year effects of the GLM are returned in the attribute "glm_year_effect", attr(x, "glm_year_effect"): log effect relative to the first year of each group of linked years (LOG_EFFECT), with a quasi-Poisson standard error conditional on the site totals (SE). This is a model-based annual trend from the counts, a quick check of the collated index, but its SE ignores the variation among sites captured by the bootstrap.

See also

flight_curve

Author

Reto Schmucki - retoshm@ceh.ac.uk