Multivariate receptor modeling from a statistical science viewpoint /

Receptor modeling is a collection of methods used to model

Bibliographic Details
Main Author: Park, Eun Sug , 1967-
Format: Thesis Book
Language:English
Published: [Place of publication not identified] : [publisher not identified] ; 1997.
Subjects:
Online Access:http://proxy.library.tamu.edu/login?url=http://proquest.umi.com/pqdweb?did=736824721&sid=1&Fmt=2&clientId=2945&RQT=309&VName=PQD
Description
Summary:Receptor modeling is a collection of methods used to model
the air pollution data. Although the main interest is
identifying pollution sources, the determination of the
number of major sources is the first problem that we
should overcome. The NUMFACT algorithm is a new method of
determining the number of underlying factors in a
multivariate system. Although NUMFACT has not been
previously described, it has been used in high profile
studies. The asymptotic distribution of NUMFACT
statistics and their associated cutoff points are derived.
Modified NUMFACT statistics and related statistics, W, are
also presented with a new decision rule for determining
the number of sources. The simulation study shows that
for the lognormal error case, they are superior to the
traditional methods such as Bartlett's statistic or the
rule of thumb methods, and for the normal error case, they
are competitive with the best of the traditional methods.
Many air pollution datasets typically consist of the
measurements on fifty or sixty variables, and it is often
too large to handle all at once. We develop two new
algorithms, SPECIESA and SPECIESB, for choosing the
species used in final model fitting. The simulation
results show that these algorithms work very well in
choosing the species generated by the sources not by the
measurement errors. Estimating the source profiles and
their contributions are our primary concerns in receptor
modeling. We take the constrained nonlinear least squares
approach to provide those estimates under as little
assumptions as possible about the model parameters. The
assumptions that we make are necessary to get the model
identifiability. A set of algorithms, VERTEX, to find the
least squares solution under different versions of the
model is introduced. The resulting estimators are shown
to be consistent and asymptotically normal under
appropriate identifiability conditions. The proposed
methods are applied to simulated data and real air
pollution data.
Item Description:Vita.
"Major Subject: Statistics".
Physical Description:xi, 153 leaves : illustrations ; 28 cm.
Issued also on microfiche from University Microfilms Inc.
Bibliography:Includes bibliographical references: pages 143-147.