Regression-tree modeling of desert tortoise habitat in the central Mojave Desert

This paper describes an interdisciplinary study of the habitat requirements of threatened desert tortoises (Gopherus agassizii) on eight 225-ha study plots in a 14000 ha study area near the southern boundary of the U.S. Army' s National Training Center at Fort Irwin in the central Mojave Desert of so uthern California. The objective of the study was to produce an empirical, statistical, GIS-based model of desert tortoise habitat use based on a comb ination of field data and data derived from various spatial databases, incl uding satellite imagery. A total of 11 primary and secondary data layers co nstitute the spatial database used for this project. Vegetation and tortois e relative density data were obtained from field surveys. Regression-tree m ethods were used to develop the statistical model. The toe has 11 terminal nodes and a residual mean deviance of 1.985, Out of 73 potential predictors in the model specification, only eight were selected by the algorithm to b e used in construction of the tree. The model suggests that tortoises tend to occur on southwest exposures and loamy soils, and that they avoid stony soils, north exposure, and areas of very low plant cover. Our results imply that soil composition and parent materials can be important determinants o f habitat suitability for desert tortoises, and for burrowing animals in ge neral. Our study also provides an example of hew the interdisciplinary inte gration of biology, earth sciences, GIS, and statistical modeling can revea l important aspects of the habitat requirements of endangered species.