repAssess: Assess sample representativeness

In track2KBA: Identifying Important Areas from Animal Tracking Data

Assess sample representativeness

Description

repAssess estimates the degree to which the space use of a tracked sample of animals represents that of the larger population.

Usage

repAssess(
  tracks,
  KDE = NULL,
  iteration = 1,
  levelUD,
  avgMethod = "mean",
  nCores = 1,
  bootTable = FALSE
)

Arguments

tracks	SpatialPointsDataFrame of spatially projected animal relocations. Must include 'ID' field.
KDE	Kernel Density Estimates for individual animals. Several input options: an estUDm, a SpatialPixels/GridDataFrame, or a RasterStack. If estUDm, must be as created by estSpaceUse or adehabitatHR::kernelUD, if Spatial* each column should correspond to the Utilization Distribution of a single individual or track. If a RasterStack, each layer must be an individual UD.
iteration	numeric. Number of times to repeat sub-sampling procedure. The higher the iterations, the more robust the result.
levelUD	numeric. Specify which contour of the utilization distribution (KDE) you wish to filter to (e.g. core area=50, home range=95).
avgMethod	character. Choose whether to use the arithmetic or weighted mean when combining individual IDs. Options are :'mean' arithmetic mean, or 'weighted', which weights each UD by the numner of points per level of ID.
nCores	numeric. The number of processing cores to use. For heavy operations, the higher the faster. NOTE: CRAN sets a maximum at 2 cores. If using the git-hub version of the package, this can be set to a maximum of one fewer than the maximum cores in your computer.
bootTable	logical (TRUE/FALSE). If TRUE, output is a list, containing in the first slot the representativeness results summarized in a table, and in the second the full results of the iterated inclusion calculations.

Details

Representativeness is assessed by fitting statistical model to the relationship between sample size and inclusion rate. Incusion rate is the proportion of out-sample points included in in-sample space use areas.

First, the set of IDs is iteratively sub-sampled, and in each iteration a set of individual Utilization Distributions (UD, 'KDE' argument) are pooled and the points of the un-selected (out-sample) IDs are overlaid on the area ('levelUD') of the UD. The proportion of these outsample points which overlap the pooled UD area is known as the inclusion rate, and represents an estimate of representativeness at each sample size. Then, a non-linear function is fit to the relationship between the inclusion rate and sample size (i.e. number of tracks/animals) in order to estimate the point at which the relationship reaches an asymptote (i.e. no more information added per new track). repAssess then estimates the representativeness of the sample by dividing the inclusion rate estimated at the maximum sample size minus 3 (for samples where n < 20), 2 (for samples < 50) or 1 (for sample >100) by this asymptote. The maximum sample size appearing in the plot will be different than the true 'n' of the dataset in order to account for the possible number of combinations of individuals, thereby ensuring a robust result. The maximum sample size reflects the number of KDEs, so if any ID has fewer than 5 points, this ID is omitted from the analysis. Finally, using this relationship, minimum representative sample sizes (70

repAssess accepts UDs calculated outside of track2KBA, if they have been converted to class RasterStack or SpatialPixelsDataFrame. However, one must make sure that the cell values represent continuous probability densities (i.e. values >=0 which integrate to 1 over the raster) and not not discrete probability masses (i.e. values >=0 which sum to 1), nor home range quantiles (i.e. 0-1, or 0-100 representing

When setting avgMethod care must be taken. If the number of points differ greatly among individuals and the UDs are calculated as classic KDEs (e.g. from estSpaceUse) then the weighted mean is likely the optimal way to pool individual UDs. However, if any other method (for example AKDE, auto-correlated KDE) was used to estimate UDs, then the arithmetic mean is the safer option.

NOTE: this function does not work with fewer than 4 IDs (tracks or individual animals).

Value

if bootTable=FALSE (the default) A single-row data.frame is returned, with columns 'SampleSize' signifying the sample size (i.e., number of KDEs)'out' signifying the percent representativeness of the sample,'type' is the type of asymptote value used to calculate the 'out' value, and 'asym' is the asymptote value used. If bootTable=TRUE, a list returned with above dataframe in first slot and full iteration results in second slot.

There are two potential values for 'type':'asymptote' is the ideal, where the asymptote value is calculated from the parameter estimates of the successful nls model fit. 'inclusion' is used if the nls fails to converge, or if the fit model is flipped and the asymptote value is negative. In these casess, the mean inclusion rate is taken for the largest sample size.'Rep70' signifies the sample size which is ~70 representative, and 'Rep95' signifies the sample size which approahces the asymptote.

Examples

library(dplyr)
tracks_raw <- track2KBA::boobies
## format data
tracks_formatted <- formatFields(
  dataGroup = tracks_raw,
  fieldID   = "track_id",
  fieldLat  ="latitude",
  fieldLon  ="longitude",
  fieldDate ="date_gmt",
  fieldTime ="time_gmt"
)

## project dataset
tracks_prj <- projectTracks(
  tracks_formatted,
  projType = "azim",
  custom = "TRUE"
)
KDE <- track2KBA::KDE_example

result <- repAssess(tracks_prj, KDE, levelUD = 50, iteration = 1)

track2KBA documentation built on Sept. 27, 2023, 5:08 p.m.