Robert M. Dorazio
Southeast Ecological Science Center
U.S. Geological Survey

TITLE:  Estimating the Geographic Distribution of a Species Using
Presence-only Records

ABSTRACT:
A maximum likelihood estimator is developed for predicting the
geographic distribution of species occurrence probabilities using
covariate measurements taken in a background sample of locations,
where the species may be present or absent, and in a presence-only
sample of locations, where the species is observed. The estimator is
consistent for the parameters of species occurrence probability
provided the background and presence-only samples are selected without
bias and provided the covariates of species occurrence and species
detection probabilities are distinct and independently distributed.
Unlike other species distribution models (e.g., Maxent), this approach
does not require species prevalence over the study area to be known.
Simulations were used to illustrate the operating characteristics of
the estimator under different sample sizes and to make comparisons
with site-occupancy estimates of occurrence based on presence-absence
data.  A geographic survey of avian species was used to compare site-
occupancy estimates of occurrence probabilities, which account for
errors in detection, with estimates of occurrence probabilities based
on the presence-only records of each species.  The estimated effects
of habitat covariates on species occurrence were quite similar for
these two classes of models; however, the presence-only-based
estimates of intercept parameters were lower and considerably less
precise than site-occupancy-based estimates of these parameters.