extractBhatt: Sample covariates according using a vector of three...

In SEEG-Oxford/seegSDM: Streamlined Functions for Species Distribution Modelling in the SEEG Research Group

Description

Generate pseudo-absence and pseudo-presence points according the procedure described by Bhatt et al. (2013) using the parameters: na, np and mu (specified by pars) and extract covariate values into a dataframe for modelling.

Where n is the number of occurrence points, na * n pseudo-absence points and np * n pseudo-presence points are generated at a maximum distance of mu decimal degrees from the occurrence points occurrence. The selection of pseudo-absence/presence points is spatially weighted based on the evidence consensus layer consensus (with values ranging from -100 to 100). Pseudo-absence points are more likely to be selected from cells with low consensus values (and never from cells with consensus 100). Pseudo-presence points are more likely to be from cells with high consensus values, and never from cells with values below threshold.

The covariate values for occurrence, pseudo-absence and pseudo-presence points are extracted from covariates and combined into a dataframe ready for modelling. Covariate values for polygon records are extracted using extractAdmin and summarized either by their mean or mode for continuous and discrete variables respectively. Discrete variables should by identified using the factors argument which will also coerce the covariates into factors in the resulting dataframe. If points = TRUE a list is returned containing the dataframe and the pseudo-presence and pseudo-absence points. If return_points = FALSE only the dataframe is returned.

Usage

extractBhatt(pars, occurrence, covariates, consensus, admin, threshold = -25,
             factor = rep(FALSE, nlayers(covariates)), return_points = FALSE, ...)

Arguments

pars	Vector of length three giving parameters for selection of pseudo-presence and pseudo-absence points. The parameters are: pars[1]: the number of pseudo-absences to select for each occurrence point (na in Bhatt et al.) pars[2]: the number of pseudo-presences to select for each occurrence point (np in Bhatt et al.) pars[3]: the maximum distance (in decimal degrees) from occurrence points at which to select pseudo-presences and pseudo-absences (mu in Bhatt et al.)
occurrence	A SpatialPointsDataFrame object giving occurrence records and containing at least the columns Admin (giving the admin level for polygons or -999 for points) and GAUL (giving the GAUL codes for polygons or NA for points).
covariates	A Raster* object giving covariates to use in the model.
consensus	A RasterLayer object giving the evidence consensus values (between -100 and 100).
admin	A RasterBrick or RasterStack object with four layers giving the GAUL codes for different admin levels. The layers must be in the order 0, 1, 2, 3, as they are in the example object admin.
threshold	A threshold evidence consensus value below which pseudo-presence points will not be selected.
factor	A logical vector stating whether each layer in covariates is discrete and should therefore be coerced into a factor.
return_points	Whether to return a list containing the dataframe and SpatialPoints objects of the pseudo-presence and pseudo-absence data (if return_points = TRUE) or just the dataframe (if return_points = FALSE).
...	Additional arguments to pass to bgSample. Note that prob is already set as TRUE, so will be ignored.

Value

If return_points = TRUE, a list with three elements:

data	A dataframe with: column 1 giving 1s for occurrence and pseudo-presence records and 0s for pseudo-absence records; columns 2 and 3 giving the coordinates of these records; and the following columns containing the extracted covariates.
pseudo_absence	A SpatialPoints object of the pseudo-absence points selected. If pars[1] was 0 or lower this will be NULL
pseudo_presence	A SpatialPoints object of the pseudo-presence points selected. If pars[2] was 0 or lower this will be NULL

Otherwise just data.

References

Bhatt et al. (2013) The global distribution and burden of dengue. Nature http://www.nature.com/nature/journal/v496/n7446/full/nature12060.html

See Also

bgDistance, bgSample

Examples

# load fake occurrence points, covariates and evidence consensus layer
data(occurrence)
data(covariates)
data(consensus)
data(admin)

# run checks, add GAUL codes and return a SpatialPoints object
occurrence <- checkOccurrence(occurrence, consensus, admin)

# run extractBhatt with:
# pseudo-absence ratio of 10:1
# pseudo-presence ratio of 0.1:1
# distance of 5 decimal degrees
# and specifying the last covariate as a discrete variable

lis <- extractBhatt(c(2, 1, 5),
                    occurrence,
                    covariates,
                    consensus,
                    admin,
                    return_points = TRUE,
                    factor = c(FALSE, FALSE, TRUE))

# look at what's produced
str(lis)

# specifically the dataframe
summary(lis$data)

# plot the true and pseudo points

# evidence consensus layer as a background
plot(consensus)
# add the pseudo-absences in light blue (note they aren't in high scoring
# consensus regions)
points(lis$pseudo_absence, pch = 16, col = 'light blue')
# and pseudo-presences in purple (not in the very low scoring regions) 
points(lis$pseudo_presence, pch = 16, col = 'purple')
# and add the occurrence points
points(occurrence, pch = 16)

SEEG-Oxford/seegSDM documentation built on May 9, 2019, 11:08 a.m.