Environmental variables
To investigate correlations between the distribution of Cochemiea
halei and its environment, we chose 19 energy and precipitation
variables from WorldClim V. 2.0, averages from 1970 to 2000, at 30
arcsec resolution (Fick & Hijmans, 2017). Soil type was determined
during field surveys using a Munsell soil identification color scale
(Munsell Color, Grand Rapids, MI), categorizing soils into ultramafic
(2.5Y hue with various color values and chroma) versus either
“non-serpentine,” (approximately 7.5YR to 10YR), or sand (Roberts,
1980). Dense sampling of occurrences of C. halei with soil type
data was performed in order to reduce error when interpolating for
missing values (Carl & Kühn, 2007; Dormann & McPherson, 2007; Dormann
et al. 2013). The soil type data from the field was mapped onto zones of
ultramafic versus non-ultramafic substrate, as indicated in the
geological map of Isla Magdalena and Isla Margarita by Rangin (1978).
The soil type raster was generated using inverse distance weighted
interpolation (Gonçalves, 2006; Grunwald, 2009) and improved using root
mean squared error and 5-fold cross validation (Gonçalves, 2006).
Four representative concentration pathways (RCPs) were used in climate
change projections: 2.6, representing the best case future concentration
of carbon in the atmosphere, through intermediate levels 4.5 and 6.0, to
the worst case scenario of 8.5, as outlined in the Intergovernmental
Panel on Climate Change’s Fifth Assessment Report (IPCC 2013; Liddicoat,
Jones, & Robertson, 2013). The climate data itself was derived from two
general circulation models (GCMs). The GCMs used were the Hadley Center
Global Environmental Model version 2-ES (HadGEM2-ES) and the Community
Climate System Model v. 4 (CCSM4), both of which are frequently used in
studies of climate change effects on habitat suitability (e.g., Bellouin
et al., 2011; Leclère et al. 2014; McQuillan & Rice, 2015; Albuquerque
et al., 2018). The HadGEM2-ES model scenarios include projections of
changes in ocean temperature and sea ice, and are especially recommended
for use in predicting changes in coastal habitat (Collins et al. 2008,
Caesar et al., 2013).