The goal of envSDM is to help automate the preparation, tuning and prediction of species distribution models. envSDM attempts to make decisions at each of these steps that are robust(ish) for running SDMs for many, many taxa.
If you are looking for packages to run species distribution models (or ecological niche models) there are plenty of better packages to choose from:
envSDM assumes that your are trying to run many, many taxa, thus there is no option to run a single taxa in parallel. The functions are all designed around the potential to run many taxa in parallel (assuming each taxa is run on a single core). For the long running functions, there is the option to return either: the object, or the path to an .rds file into which the object is saved.
Preparation includes generating:
- generation of a (possibly buffered) minimum convex polygon around presences to limit the rest of the process (predict boundary)
- density raster of presences
- spatially thickened (Vollering et al. 2019) background points against density raster
- balanced spatial folds from the presences and background points
- environmental data for presences and background points
- ensuring the environmental variables used are not correlated beyond a threshold, per taxa
Tuning includes:
- three possible algorithms:
-
randomForest::randomForest()- always using the
randomForest()sampsizeargument downsample to the minimum number of presences
- always using the
maxnet::maxnet()predicts::envelope()
-
- ability to use multiple metrics for choosing a ‘best’ tune
Prediction includes:
- only predicting to a predict boundary established during the preparation
- threshold to maximum of specificity + sensitivity
Installation
envSDM is not on CRAN.
You can install the development version from GitHub with:
# install.packages("devtools")
devtools::install_github("dew-landscapes/envSDM")