Similar to others, this engine enables the fitting and prediction of
log-Gaussian Cox process (LGCP) and Inhomogeneous Poisson process (IPP) processes.
It uses the scampr package, which uses maximum likelihood estimation
fitted via TMB (Template Model Builder).
It also support the addition of spatial latent effects which can be added via Gaussian fields and approximated by 'FRK' (Fixed Rank Kriging) and are integrated out using either variational or Laplace approximation.
The main use case for this engine is as an alternative to engine_inlabru() and
engine_inla() for fitting iSDMs, e.g. those combining both presence-only
and presence-absence point occurrence data.
Arguments
- x
distribution()(i.e.BiodiversityDistribution) object.- type
The mode used for creating (posterior or prior) predictions. Either stting
"link"or"response"(Default:"response").- dens
A
characteron how predictions are made, either from the"posterior"(Default) or"prior".- maxit
A
numericon the number of iterations for the optimizer (Default:500).- ...
Other parameters passed on.
Value
An Engine.
Details
This engine may only be used to predict for one or two datasets at most. It supports only presence-only PPMs and presence/absence Binary GLMs, or 'IDM' (for an integrated data model).
Note
The package can currently be installed from github directly only
"ElliotDovers/scampr"Presence-absence models in SCAMPR currently only support cloglog link functions!
References
Dovers, E., Popovic, G. C., & Warton, D. I. (2024). A fast method for fitting integrated species distribution models. Methods in Ecology and Evolution, 15(1), 191-203.
Dovers, E., Stoklosa, D., and Warton D. I. (2024). Fitting log-Gaussian Cox processes using generalized additive model software. The American Statistician, 1-17.
See also
Other engine:
engine_bart(),
engine_breg(),
engine_gdb(),
engine_glm(),
engine_glmnet(),
engine_inla(),
engine_inlabru(),
engine_stan(),
engine_xgboost()
Examples
if (FALSE) { # \dontrun{
# Load background
background <- terra::rast(system.file('extdata/europegrid_50km.tif',
package='ibis.iSDM',mustWork = TRUE))
# Add GLM as an engine
x <- distribution(background) |> engine_scampr()
} # }