Department of
presents
Raymond Webster, Jungsoon Choi, Jiajun Liu
Modeling Observer Effects on Animal Density and Detection for Combined Distance and Capture-Recapture Data
Abstract: The estimation of animal density using distance
sampling methods generally ignores the effects of survey transects or
observers on local animal density or detection probability. Typically,
density is assumed to be uniform around the transect, which fails to
allow for avoidance of or attraction to the transect by an animal
species, while detection probability must be perfect on the transect,
which may be untrue for many species. We propose parametric models for
combined distance and capture-recapture survey data from line and point
transect surveys that allow for two types of movement: permanent
avoidance of or attraction to a transect, and temporary displacement of
animals in the vicinity of a transect. Through a simulation study, we
show that, provided sufficient animals are detected, the parameter
estimators have little bias, and can lead to improved density estimates
over the uniform model. We apply our models to a point transect survey
of birds in the Great Smoky Mountains National Park.
*This is joint work with
Multivariate Spatiotemporal Model for Speciated Fine Particulate Matter*
Abstract: Fine particle matter (PM2.5) is an atmospheric
pollutant that has been linked to serious health problems, including
mortality. PM2.5 is a mixture of pollutants, and it has five main
components: sulfate, nitrate, total carbonaceous mass, ammonium, and
crustal materials. These components have complex spatial -temporal
dependency and cross-dependency structures. It is important to gain
insight and better understanding about the spatial distribution of each
component of total PM2.5, and also to estimate how the composition of
PM2.5 might change with location and season.
In this work we introduce a multivariate spatiotemporal model for
speciated PM2.5. We propose a Bayesian hierarchical framework with
spatiotemporally varying coefficients. In addition, a linear
coregionalization model is developed to account for spatial and
temporal dependence across the five different components as well as the
associations among components at a given location and time. We apply
our model to speciated PM2.5 monitoring data in the United States for
the year 2004 from the U.S. Environmental Protection Agency.
*This is joint work with
Analysis of Gene Expression Data Using the Gene Ontology*
Abstract: New technologies for biological systems give scientists the
ability to measure thousands of gene expression for bio-molecule including
genes, proteins, lipids and metabolites. Our goal is to use domain knowledge,
e.g. the Gene Ontology, to guide analysis of such data. By focusing on
domain-aggregated results at, say the molecular function level, increased interpretability
is available to biological scientists beyond what is possible if results are
presented at the gene level. We use a “top-down” approach to
perform domain aggregation by first combining gene expression before testing
for differentially expressed patterns. This is contrast to the more standard
“bottom-up approach where genes are first tested individually then
aggregated by domain knowledge. Our method is assessed and compared to other
methods using a series of simulation studies. Implications from analysis of a
real dataset are also presented.
*This is joint work with J. M. Hughes-Oliver
(NCSU) and Alan Menius (GSK)
Friday, August
25, 2006
3:35 - 4:35 pm
206 Cox Hall
Refreshments will be served on the second floor of Dabney Hall (left of Room 222) at 3:00 pm.