Partition the study area into rectangular tiles

Source: R/sperrorest_resampling.R

partition_tiles divides the study area into a specified number of rectangular tiles. Optionally small partitions can be merged with adjacent tiles to achieve a minimum number or percentage of samples in each tile.

partition_tiles(
  data,
  coords = c("x", "y"),
  dsplit = NULL,
  nsplit = NULL,
  rotation = c("none", "random", "user"),
  user_rotation,
  offset = c("none", "random", "user"),
  user_offset,
  reassign = TRUE,
  min_frac = 0.025,
  min_n = 5,
  iterate = 1,
  return_factor = FALSE,
  repetition = 1,
  seed1 = NULL
)

Arguments

data

data.frame containing at least the columns specified by coords

coords

vector of length 2 defining the variables in data that contain the x and y coordinates of sample locations

dsplit

optional vector of length 2: equidistance of splits in (possibly rotated) x direction (dsplit[1]) and y direction (dsplit[2]) used to define tiles. If dsplit is of length 1, its value is recycled. Either dsplit or nsplit must be specified.

nsplit

optional vector of length 2: number of splits in (possibly rotated) x direction (nsplit[1]) and y direction (nsplit[2]) used to define tiles. If nsplit is of length 1, its value is recycled.

rotation

indicates whether and how the rectangular grid should be rotated; random rotation is only between -45 and +45 degrees.

user_rotation

if rotation='user', angles (in degrees) by which the rectangular grid is to be rotated in each repetition. Either a vector of same length as repetition, or a single number that will be replicated length(repetition) times.

offset

indicates whether and how the rectangular grid should be shifted by an offset.

user_offset

if offset='user', a list (or vector) of two components specifying a shift of the rectangular grid in (possibly rotated) x and y direction. The offset values are relative values, a value of 0.5 resulting in a one-half tile shift towards the left, or upward. If this is a list, its first (second) component refers to the rotated x (y) direction, and both components must have same length as repetition (or length 1). If a vector of length 2 (or list components have length 1), the two values will be interpreted as relative shifts in (rotated) x and y direction, respectively, and will therefore be recycled as needed (length(repetition) times each).

reassign

logical (default TRUE): if TRUE, 'small' tiles (as per min_frac and min_n arguments and get_small_tiles) are merged with (smallest) adjacent tiles. If FALSE, small tiles are 'eliminated', i.e. set to NA.

min_frac

numeric >=0, <1: minimum relative size of partition as percentage of sample; argument passed to get_small_tiles. Will be ignored if NULL.

min_n

integer >=0: minimum number of samples per partition; argument passed to get_small_tiles. Will be ignored if NULL.

iterate

argument to be passed to tile_neighbors

return_factor

if FALSE (default), return a represampling object; if TRUE (used internally by other sperrorest functions), return a list containing factor vectors (see Value)

repetition

numeric vector: cross-validation repetitions to be generated. Note that this is not the number of repetitions, but the indices of these repetitions. E.g., use repetition = c(1:100) to obtain (the 'first') 100 repetitions, and repetition = c(101:200) to obtain a different set of 100 repetitions.

seed1

seed1+i is the random seed that will be used by set.seed in repetition i (i in repetition) to initialize the random number generator before sampling from the data set.

Value

A represampling object. Contains length(repetition) resampling objects as repetitions. The exact number of folds / test-set tiles within each resampling objects depends on the spatial configuration of the data set and possible cleaning steps (see min_frac, min_n).

Note

Default parameter settings may change in future releases. This function, especially the rotation and shifting part of it and the algorithm for cleaning up small tiles is still a bit experimental. Use with caution. For non-zero offsets (offset!='none')), the number of tiles may actually be greater than nsplit[1]*nsplit[2] because of fractional tiles lurking into the study region. reassign=TRUE with suitable thresholds is therefore recommended for non-zero (including random) offsets.

Examples

data(ecuador)
set.seed(42)
parti <- partition_tiles(ecuador, nsplit = c(4, 3), reassign = FALSE)
# plot(parti,ecuador)
# tile A4 has only 55 samples
# same partitioning, but now merge tiles with less than 100 samples to
# adjacent tiles:
parti2 <- partition_tiles(ecuador,
  nsplit = c(4, 3), reassign = TRUE,
  min_n = 100
)
# plot(parti2,ecuador)
summary(parti2)
#> $`1`
#>       n.train n.test
#> X1:Y3     600    151
#> X2:Y2     626    125
#> X3:Y1     584    167
#> X3:Y2     574    177
#> X3:Y3     620    131
#> 
# tile B4 (in 'parti') was smaller than A3, therefore A4 was merged with B4,
# not with A3
# now with random rotation and offset, and tiles of 2000 m length:
parti3 <- partition_tiles(ecuador,
  dsplit = 2000, offset = "random",
  rotation = "random", reassign = TRUE, min_n = 100
)
# plot(parti3, ecuador)
summary(parti3)
#> $`1`
#>       n.train n.test
#> X2:Y2     452    299
#> X3:Y1     562    189
#> X3:Y2     488    263
#>