The hardware and bandwidth for this mirror is donated by METANET, the Webhosting and Full Service-Cloud Provider.
If you wish to report a bug, or if you are interested in having us mirror your free-software or open-source project, please feel free to contact us at mirror[@]metanet.ch.
This vignette illustrates the various functions of PointedSDMs by using three datasets of the solitary tinamou (Tinamus solitarius) – a species of ground bird found on the eastern side of Brazil. Due to package dependencies, this vignette is not run. However the data and R script are available such that the user may carry out inference.
Firstly, we load in the datasets and objects required for this
vignette. The SolitaryTinamou dataset attached to this package
contains a list
of four objects; for ease of use, we make
new objects for the items in this list
.
data('SolitaryTinamou')
projection <- "+proj=longlat +ellps=WGS84"
covariates <- terra::rast(system.file('extdata/SolitaryTinamouCovariates.tif',
package = "PointedSDMs"))
datasets <- SolitaryTinamou$datasets
region <- SolitaryTinamou$region
mesh <- SolitaryTinamou$mesh
The first item is a list
of three datasets:
eBird, Gbif and Parks. The first two are
data.frame
objects containing only two variables:
X
and Y
describing the latitude and longitude
coordinates of the species location respectively. As a result of this,
these two datasets are considered to be present only datasets
in our integrated model.
The other dataset (Parks) is also a data.frame
object. It contains the two coordinate variables present in the first
two datasets, but contains two additional variables:
Present
, a binary variable describing the presence
(1) or absence (0) of the species at the given
coordinates, and area
describing the area of the park.
Since we have information on the presences and absences of the species
in this dataset, we consider it a presence absence dataset.
Region
is a sf
object which give the
boundary of the park containing the species; it was used in the
mesh construction and for the plots in this vignette.
The next object is covariates
, a spatRaster
objects of the covariates (Forest, NPP and Altitude)
describing the area of the parks. We stack these three objects together
using the stack
function, and then scale them.
Finally we require a Delaunay triangulated mesh for the construction of the spatial field. A plot of the mesh used for this vignette is provided below.
To set up an integrated species distribution model with
PointedSDMs
, we initialize it with the
startISDM
function – which results in an R6
objects with additional slot functions to further customize the model.
The base model we run for these data comprises of the spatial covariates
and an intercept term for each dataset.
base <- startISDM(datasets, spatialCovariates = covariates,
Projection = projection, responsePA = 'Present', Offset = 'area',
Mesh = mesh, pointsSpatial = NULL)
Using the .$plot
function produces a gg object
of the points used in this analysis by dataset; from this plot, we see
that most of the species locations are found towards the eastern and
south-central part of the park.
base$plot(Boundary = FALSE) +
geom_sf(data = st_boundary(region)) +
ggtitle('Plot of the species locations by dataset')
In this model, we also include prior information for the
Forest effect using $priorsFixed
.
To run the integrated model, we use the fitISDM
function
with the data
argument as the object created with the
startISDM
function above.
If we would like to correlate the spatial fields across the datasets
, we can specify pointsSpatial = 'correlate'
in
startISDM()
:
If we wanted to make predictions of the shared spatial random field
across the map, we set spatial = TRUE
in the generic
predict
function.
spatial_predictions <- predict(fieldsModel, mesh = mesh,
mask = region,
spatial = TRUE,
fun = 'linear')
And subsequently plot using the generic plot
function.
However if we wanted to make predictions of the bias field, we would
do this by setting biasfield = TRUE
.
The last function of interest is datasetOut
, which
removes a dataset out of the full model, and then calculates a
cross-validation score with this reduced model. In this case, we try the
function out by removing the eBird dataset.
These binaries (installable software) and packages are in development.
They may not be fully stable and should be used with caution. We make no claims about them.