cc_outl: Identify Geographic Outliers in Species Distributions

Description

Removes out or flags records that are outliers in geographic space according to the method defined via the method argument. Geographic outliers often represent erroneous coordinates, for example due to data entry errors, imprecise geo-references, individuals in horticulture/captivity.

Usage

cc_outl(
  x,
  lon = "decimalLongitude",
  lat = "decimalLatitude",
  species = "species",
  method = "quantile",
  mltpl = 5,
  tdi = 1000,
  value = "clean",
  sampling_thresh = 0,
  verbose = TRUE,
  min_occs = 7,
  thinning = FALSE,
  thinning_res = 0.5
)

Value

Depending on the ‘value’ argument, either a data.frame

containing the records considered correct by the test (“clean”) or a logical vector (“flagged”), with TRUE = test passed and FALSE = test failed/potentially problematic . Default = “clean”.

Arguments

x: data.frame. Containing geographical coordinates and species names.
lon: character string. The column with the longitude coordinates. Default = “decimalLongitude”.
lat: character string. The column with the latitude coordinates. Default = “decimalLatitude”.
species: character string. The column with the species name. Default = “species”.
method: character string. Defining the method for outlier selection. See details. One of “distance”, “quantile”, “mad”. Default = “quantile”.
mltpl: numeric. The multiplier of the interquartile range (method == 'quantile') or median absolute deviation (method == 'mad')to identify outliers. See details. Default = 5.
tdi: numeric. The minimum absolute distance (method == 'distance') of a record to all other records of a species to be identified as outlier, in km. See details. Default = 1000.
value: character string. Defining the output value. See value.
sampling_thresh: numeric. Cut off threshold for the sampling correction. Indicates the quantile of sampling in which outliers should be ignored. For instance, if sampling_thresh == 0.25, records in the 25 (no sampling correction).
verbose: logical. If TRUE reports the name of the test and the number of records flagged.
min_occs: Minimum number of geographically unique datapoints needed for a species to be tested. This is necessary for reliable outlier estimation. Species with fewer than min_occs records will not be tested and the output value will be 'TRUE'. Default is to 7. If method == 'distance', consider a lower threshold.
thinning: forces a raster approximation for the distance calculation. This is routinely used for species with more than 10,000 records for computational reasons, but can be enforced for smaller datasets, which is recommended when sampling is very uneven.
thinning_res: The resolution for the spatial thinning in decimal degrees. Default = 0.5.

Details

The method for outlier identification depends on the method argument. If “quantile”: a boxplot method is used and records are flagged as outliers if their mean distance to all other records of the same species is larger than mltpl * the interquartile range of the mean distance of all records of this species. If “mad”: the median absolute deviation is used. In this case a record is flagged as outlier, if the mean distance to all other records of the same species is larger than the median of the mean distance of all points plus/minus the mad of the mean distances of all records of the species * mltpl. If “distance”: records are flagged as outliers, if the minimum distance to the next record of the species is > tdi. For species with records from > 10000 unique locations a random sample of 1000 records is used for the distance matrix calculation. The test skips species with fewer than min_occs, geographically unique records.

The likelihood of occurrence records being erroneous outliers is linked to the sampling effort in any given location. To account for this, the sampling_cor option fetches the number of occurrence records available from www.gbif.org, per country as a proxy of sampling effort. The outlier test (the mean distance) for each records is than weighted by the log transformed number of records per square kilometre in this country. See for https://besjournals.onlinelibrary.wiley.com/doi/full/10.1111/2041-210X.13152 an example and further explanation of the outlier test.

Examples

Run this code


x <- data.frame(species = letters[1:10],
                decimalLongitude = runif(100, -180, 180),
                decimalLatitude = runif(100, -90,90))

cc_outl(x)
cc_outl(x, method = "quantile", value = "flagged")
cc_outl(x, method = "distance", value = "flagged", tdi = 10000)
cc_outl(x, method = "distance", value = "flagged", tdi = 1000)