Skip to contents

Prepare landscape rasters for focal statistics via Python

Source: R/python_focal_prep.R
python_focal_prep.Rd

Prepare for running focal statistics on a landscape raster via Python, to generate inputs for use with species distribution models.

Usage

python_focal_prep(
  x,
  SDM,
  classified = FALSE,
  fill = TRUE,
  suffix = NULL,
  pixel_value = NULL,
  subset = NULL,
  dir = NULL,
  overwrite = FALSE,
  ...
)

Arguments

x

SpatRaster with layer names corresponding to landscape_name

SDM

The name of intended species distribution model, for which x will be reclassified: "riparian", "waterbird_fall", "waterbird_win", or "tima"

classified

logical; passed to create_predictor_stack()

fill

logical; passed to create_predictor_stack()

suffix

Character string; custom suffix appended to layer names (optional unless mask is not NULL); see Details.

pixel_value

Optional numeric value to use in place of 1 where land covers are present; default NULL

subset

Optional SpatRaster or string representing filepath to a raster. See Details.

dir

Optional string representing directory passed to terra::writeRaster(), as (dir/SDM/landscape_name). See Details.

...

additional arguments passed to terra::writeRaster()

Value

SpatRaster, though primarily used to write layers to file for use with python_focal_stats()

Details

This is a wrapper function that calls create_predictor_stack() to split a land cover raster into separate layers representing the presence (1) or absence (0) of each required land cover class predictor for use in calculating focal statistics. The input raster should already be encoded with the land cover classes listed in the key. To help with creating such a raster, see classify_landcover.sf() to map land cover polygons to the land cover classes in the key. See documentation of create_predictor_stack() and classify_landcover.SpatRaster() for information about Warning messages.

If suffix is provided, it is appended to the layer name. If pixel_value is provided, values representing land cover presence (1) are replaced with this value (e.g., the area of each pixel).

If subset is provided, either as a SpatRaster or a filepath to a raster, the subset layer is masked by the presence of each land cover layer, e.g. to extract from surface water data the extent of surface water within each land cover class. In this case, both the land cover presence data and the subset data are returned; to distinguish between these results, suffix is required to have two values. The first is appended to the land cover presence data and the second is appended to the subset data. See examples.

All output can be optionally written to dir/SDM/landscape_name where landscape_name is taken from the names of the input raster(s) x.

See also

python_focal_stats()

Examples

codenums = DeltaMultipleBenefits::key$CODE_NUM
r <- terra::rast(matrix(sample(codenums, size = 1000, replace = TRUE), ncol = 100, nrow = 100))
watwin = suppressWarnings(classify_landcover(r, SDM = 'waterbird_win', verbose = FALSE))
watwin_pred = python_focal_prep(watwin, SDM = 'waterbird_win')
#> Warning: Extreme Caution Advised. Land cover classes representing a substantial proportion of the landscape are not represented by any of the predictors fot the selected SDM. Check input raster for errors.
#>    CODE_NAME count        prop
#> 1       <NA>   120 0.012820513
#> 2       <NA>    60 0.006410256
#> 3       <NA>   120 0.012820513
#> 4       <NA>   110 0.011752137
#> 5       <NA>   100 0.010683761
#> 6       <NA>   190 0.020299145
#> 7       <NA>  2930 0.313034188
#> 8       <NA>   510 0.054487179
#> 9       <NA>   170 0.018162393
#> 10      <NA>   470 0.050213675
#> 11      <NA>   220 0.023504274
#> 12      <NA>   500 0.053418803
#> 13      <NA>   110 0.011752137
#> 14      <NA>   400 0.042735043
#> 15      <NA>   270 0.028846154
#> ww grain corn field row rice fal alf ip dryp dev woodw wet duwet water barren for
#> Warning: Extreme Caution Advised. Land cover classes are missing from the input raster but are expected by the selected SDM. Check input raster for errors.
#> Because fill = TRUE, creating missing rasters with all zero values, butconfirm they are truly absent from the landscape.

# return the area of the pixel where each land cover class is present
# (useful for summing over moving windows)
watwin_area = python_focal_prep(watwin, SDM = 'waterbird_win', pixel_value = 0.09)
#> Warning: Extreme Caution Advised. Land cover classes representing a substantial proportion of the landscape are not represented by any of the predictors fot the selected SDM. Check input raster for errors.
#>    CODE_NAME count        prop
#> 1       <NA>   120 0.012820513
#> 2       <NA>    60 0.006410256
#> 3       <NA>   120 0.012820513
#> 4       <NA>   110 0.011752137
#> 5       <NA>   100 0.010683761
#> 6       <NA>   190 0.020299145
#> 7       <NA>  2930 0.313034188
#> 8       <NA>   510 0.054487179
#> 9       <NA>   170 0.018162393
#> 10      <NA>   470 0.050213675
#> 11      <NA>   220 0.023504274
#> 12      <NA>   500 0.053418803
#> 13      <NA>   110 0.011752137
#> 14      <NA>   400 0.042735043
#> 15      <NA>   270 0.028846154
#> ww grain corn field row rice fal alf ip dryp dev woodw wet duwet water barren for
#> Warning: Extreme Caution Advised. Land cover classes are missing from the input raster but are expected by the selected SDM. Check input raster for errors.
#> Because fill = TRUE, creating missing rasters with all zero values, butconfirm they are truly absent from the landscape.

# mask another raster (e.g., surface water data) by the presence of each
# land cover class:
w = watwin # simulate surface water data
levels(w) = NULL
terra::coltab(w) = NULL
terra::values(w) <- sample(c(0,1), size = 10000, replace = TRUE)
#pfld = python_focal_prep(watwin, SDM = 'waterbird_win', pixel_value = 0.09, mask = w)
#returns error because two suffixes need to be provided
pfld = python_focal_prep(watwin, SDM = 'waterbird_win', pixel_value = 0.09, mask = w,
                         suffix = c('_area', '_pfld')) # works
#> Warning: Extreme Caution Advised. Land cover classes representing a substantial proportion of the landscape are not represented by any of the predictors fot the selected SDM. Check input raster for errors.
#>    CODE_NAME count        prop
#> 1       <NA>   120 0.012820513
#> 2       <NA>    60 0.006410256
#> 3       <NA>   120 0.012820513
#> 4       <NA>   110 0.011752137
#> 5       <NA>   100 0.010683761
#> 6       <NA>   190 0.020299145
#> 7       <NA>  2930 0.313034188
#> 8       <NA>   510 0.054487179
#> 9       <NA>   170 0.018162393
#> 10      <NA>   470 0.050213675
#> 11      <NA>   220 0.023504274
#> 12      <NA>   500 0.053418803
#> 13      <NA>   110 0.011752137
#> 14      <NA>   400 0.042735043
#> 15      <NA>   270 0.028846154
#> ww grain corn field row rice fal alf ip dryp dev woodw wet duwet water barren for
#> Warning: Extreme Caution Advised. Land cover classes are missing from the input raster but are expected by the selected SDM. Check input raster for errors.
#> Because fill = TRUE, creating missing rasters with all zero values, butconfirm they are truly absent from the landscape.