Thin points so that no cell holds more than a few

Reduces the effect of uneven survey effort. Where records pile up because somewhere was visited often rather than because the species is common there, a model fitted to the raw points learns the sampling as though it were the species.

Usage

thin_points(
  x,
  coords = NULL,
  n = 1,
  cellsize = NULL,
  bins = 20,
  type = c("hex", "grid"),
  seed = 1
)

Arguments

x

A data frame of points.

coords

The two coordinate columns. Guessed when omitted.

n

Maximum number of points to keep per cell.

cellsize

Cell size in the units of the coordinates. Ignored when bins is used.

bins

Approximate number of cells across the x range, as an alternative to naming a cellsize.

type

"hex" for a hexagonal lattice, or "grid" for a square one.

seed

Random seed, since which points survive is a random choice among those sharing a cell.

Value

The rows of x that survive thinning, in their original order.

Details

Thinning is a blunt instrument and it throws data away. It is worth doing when the clustering is an artefact of where people looked, and worth not doing when the clustering is the signal – there is no way for the function to tell which, so the judgement stays with you.

The hexagonal lattice is the default for the same reason hex_bin() uses one: its cells have neighbours all at equal distance, where a square grid does not, so thinning is not subtly directional.

See also

hex_bin(), which uses the same lattice to summarise rather than to thin, and spatial_sorting_bias() for the related problem in a train/test split.

Other spatial plots: ensemble_summary(), hex_bin(), mess(), niche_equivalency(), niche_overlap(), plot.fancyfx_equivalency(), plotExtrapolation(), plotHexbin(), plotUncertainty()

Examples

set.seed(1)
# Heavily oversampled in one corner
records <- data.frame(
  x = c(runif(400, 0, 2), runif(100, 0, 10)),
  y = c(runif(400, 0, 2), runif(100, 0, 10))
)

nrow(records)
#> [1] 500
nrow(thin_points(records, n = 1, bins = 10))
#> [1] 80