This species distribution dataset depicts the occurrence
probability for Bromus tectorum (cheatgrass),which is one of four of the exotic annual grasses of most concern when it comes to fire in the
desert. These species can provide
persistent, continuous fuel beds for fires to spread, enhancing the potential
for increased fire size and frequency (Brooks and Minnich 2006).
The species distribution models were developed using MaxEnt
(Phillips and Dudik 2008, Elith et al. 2011), a map-based modeling software
that has performed well with presence-only data (Elith et al. 2006). The model
assigns a probability of species’ occurrence in each grid cell based on a
machine-learning algorithm that identifies the species’ best distribution pattern
through iterative contrasts between occurrence locations and a sample of 10 000
background points.
We downloaded the occurrence data from the Global
Biodiversity Information Facility (GBIF) (www.gbif.org),
and for each species, we ran three cross-validated model runs using ten
thousand random background points. This
resulted in a total of 95 presence records for training and 48 for testing for B. tectorum. We used hinge features, linear, and
quadratic, with an increase in regularization with a beta of 2.5, to reduce
model overfitting (Elith et al. 2011). We also used jackknife tests to estimate
predictor variable importance. We used
predictor variables hypothesized to be associated with plant distribution
patterns in desert ecosystems.
The average test AUC for the replicate runs was 0.898 for B. tectorum.