Data-driven identification of grassland types
First, a total of 49 independent grassland locations with active prairie
dog colonies within GPCA El Tokio were identified with the use of
previous literature (Ceballos et al. 1993; Trevino-Villarreal et al.
1998; Scott-Morales et al. 2004; Estrada-Castillon et al. 2010),
up-to-date Google Earth imagery and historical and present delimitations
of colonies (provided by the Mexican organizations PROFAUNA and
Organizacion Vida Silvestre A.C.-OVIS). In order to select a
representative sample of sites covering the varying environmental
conditions from these locations, a data-driven clustering approach was
used. We used a self-organizing map (SOM), a type of artificial neural
network that is trained using competitive learning and well suited to
finding clusters within data, as implemented in the package kohonen
(Wehrens and Buydens, 2007; R version 4.0, R Development Core Team
2020). Using geospatial data on mean annual precipitation, elevation,
slope, soil type and mean annual temperature (see Appendix S1 for the
specific data sets used), this analysis clustered all grassland
locations into eight groups, four of which occupy most of GPCA El Tokio
and are therefore here considered as distinct grassland types (Fig. 1a):
1) Agricultural (Agri): characterized by agricultural land use, xerosol
haplic soils, total annual precipitation between 300 to 400 mm and
temperature between 14-16 °C; 2) Arid: characterized by solonchak orthic
soils, low elevation and total annual precipitation from 200-400 mm; 3)
Calcareous (Calc): characterized by xerosol calcic soils, total annual
precipitation between 300 to 400 mm, low elevation and temperatures
between 14-16 °C and 4) Mountain (Mount): characterized by litosol, high
precipitation ranging from 400 to 500 mm, temperature between 14-16 °C
and high elevation.