Desert Tortoise - Species Distribution Model, DRECP

The Desert Tortoise (Gopherus agassizii) species distribution model includes a continuous probability surface from the USGS statistical model completed by Kenneth E. Nussear, Todd C. Esque, Richard D. Inman, Leila Gass, Kathryn A. Thomas, Cynthia S. A. Wallace, Joan B. Blainey, David M. Miller, and Robert H. Webb and a binary layer produced by the Conservation Biology Institute. The binary layer depicts predicted suitable habitat using the equal training sensitivity and specificity threshold (0.506). The binary output was also processed to exclude agriculture, developed, and disturbed areas, based on the DRECP land cover/natural vegetation dataset provided by Todd Keeler-Wolf of the CA Dept. of Fish and Game. 

The original USGS habitat model for the desert tortoise (Gopherus agassizii) was developed using an extensive set of field-collected presence data, information from literature, and expert knowledge. Sixteen environmental data layers were converted into a 1 km grid covering the study area and merged with desert tortoise presence data for input into the Maxent modeling algorithm. 

The following variables were included in the final selected model:

Mean dry season precipitation for 30-year normal period (Blainey et al. 2007)

Mean wet season precipitation for 30-year normal period (Blainey et al. 2007)

Elevation derived from 30 m NED DEM (USGS)

Average surface roughness derived from 30 m NED DEM (USGS)

Percent smooth derived from 30 m NED DEM (USGS)

Average soil bulk density (STATSGO)

Depth to bedrock (STATSGO)

Average percentage of rocks > 254 mm B-axis diameter (STATSGO)

Perennial plant cover (Remotely-sensed MODIS data)

The Maxent model provides output of the statistical probability of habitat potential and was used to produce a map of potential desert tortoise habitat for parts of the Mojave and Sonoran Deserts. This type of analysis, while robust in its predictions of habitat, does not account for anthropogenic changes that may have altered habitat with relatively high potential into areas with lower potential.

This model had a high AUC test score (0.93) and had a significant Pearson’s correlation coefficient of 0.74 (p < 0.01), indicating a substantial agreement between the predicted habitat and the observed localities of desert tortoises. The model produced output with habitat-potential scores ranging from 0 to 1, plus an area that was not estimable because environmental data were not available for one or more layers. 

For model specifics, please refer to the supporting document “Modeling Habitat of the Desert Tortoise (Gopherus agassizii) in the Mojave and Parts of the Sonoran Deserts of California, Nevada, Utah, and Arizona”, 2009.