| title | tags | authors | affiliations | date | bibliography | ||||||||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
Fauxcurrences.jl: simulation of realistic joint species distributions |
|
|
|
2022-06-30 |
paper.bib |
The analysis of species distribution from document records of observations (occurrences) is a central activity of the fields of biodiversity sciences and biogeography. The spatial distribution of species is often highly structured in space, with strong auto-correlation due to both species dispersal and the auto-correlated nature of suitable habitats. Therefore, generating pseudo-random samples of species occurrences is not a trivial task, albeit one that is emerging as a required to develop null expectations for various predictive models.
The Fauxcurrences.jl package is a rapid, memory-efficient re-implementation of
fauxcurrence, a R package by REF. It integrates with the rest of the
EcoJulia ecosystem, notably the SimpleSDMLayers.jl package for the
manipulation of geospatial data used in species distribution analysis REF,
and the GBIF.jl package to programmatically query species occurrence data from
the Global Biodiversity Infomation Facility REF. Fauxcurrences.jl is built
at an extension of the SimpleSDMLayers.jl package, and re-uses most of its
functionalities (notably to ensure that the simulated data are always located
within valued cells of the raster layer serving as a reference).
Fauxcurrences.jl was designed to be extremely memory-efficient, in order to
allow multiple runs of simulated data to proceed in parallel even on modest
hardware. In addition, compared to the original fauxcurrence package, it
relies on different measures of statistical divergence (with known bounds and no
need for data smoothing), and offer the possibility to introduce a hierarchy in
the pairwise relationship of species by assigning different weights to their
pairwise divergence in the overall scoring.
This last point is particularly important as the distribution of interacting
species is emerging as a new area of study in biogeography REF. In brief,
when species do interact, there is value in relying on null models that give
more weight to their statistical divergence than to that of pairs of species who
do not REF. To that effect, the Fauxcurrences.jl package offers various
weighting schemes for the structure of pairwise dependence between species.
The intended audience for Fauxcurrences.jl is researchers in biogeography and
macro-ecology, notably those who need to simulate realistic and large ensembles
of species observations datasets. Because the package allows to pick the size of
the simulated dataset to be generated, it is likely to be particularly useful as
a tool to help method development, by investigating which sample size is
required REF, and the extent to which differential sampling between species
is impacting the results REF. Finally, it can be incorporated in graduate
classes in species distribution analysis, as a way to generate mock datasets to
illustrate the way various species distribution models work.
We thank Owen G. Osborne and Adam A. Algar for discussions about the measure of statistical divergence.