NUSSEAR, K. E.; TRACY, C. Richard; HEATON, J. S.; MEDICA, P. A.; University of Nevada, Reno; University of Nevada, Reno; University of Redlands; U. S. Fish and Wildlife Service: An application of Neural Network modeling of desert tortoise activity to improve species monitoring.

The federally-listed desert tortoise is currently the focus of a large, multi-state monitoring program that uses distance sampling to estimate densities from transects. A critical assumption of this technique is that all animals very close to the transect are observed. Because desert tortoises spend large proportions of the year in under-ground burrows, this assumption is frequently not met. Therefore, this transect technique requires a correction factor accounting for the proportion of animals active and available to be counted, G(0), to account for the under-sampled proportion of the population. Estimating G(0) is currently accomplished by monitoring a small number of animals (N = 6-12) which are scored for behavior several times daily during the transect sampling period. Collecting these data is very costly, and lacks precision due to the small sample size. We are modeling the link between biophysical attributes of the environment and the proportions of animals active at any given week of the year, and hour of day. These models are based upon empirical observations of ~120 tortoises monitored over a three-year period to explore the influence of environmental variables on tortoise activity. The inputs to the model include: environmental temperatures, operative temperatures, rainfall, solar radiation, among others. We employ a fusion of biophysical and neural network modeling to allow for, and benefit from, the complex interactions existing among the environmental variables included in the model. We show that tortoise activity varies as a function of year, season, and time of day. We present an initial model that identifies influential environmental variables affecting activity.