L. Mark Berliner National Center for Atmospheric Research P. O. Box 3000 Boulder, CO 80307-3000
PH: 303-497-1359 FAX: 303-497-1333 berliner@ncar.ucar.edu
SESSION CHAIR AND AFFILIATION
Keith Crank, National Science Foundation
MAILING ADDRESS (Reg mail, incl zip, + phone + fax + eamail)
Dr. Keith Crank, 1025N National Science Foundation 4201 Wilson Road Arlington, VA 22203-9966
PH: 703-306-1885 kcrank@nsf.gov
SPEAKER 1:
Barbara A. Bailey, National Center for Atmospheric Research
TITLE OF PAPER:
Modeling the Spatial and Temporal Distribution of Cloud Cover
CO-AUTHORS AND AFFILIATIONS:
L. Mark Berliner, National Center for Atmospheric Research and Ohio State University
MAILING ADDRESS
Barbara Bailey National Center for Atmospheric Research P. O. Box 3000 Boulder, CO 80307-3000
PH: 303-497-1720 FAX: 303-497-1333 bbailey@ncar.ucar.edu
KEYWORDS
Climate model; Fourier coefficients; Parameterization; Satellite data
ABSTRACT
Clouds play a fundamental role in controlling the amount of solar and infrared radiation available to the climate system and therefore prediction of cloud cover amount is of great importance to climate modeling. Most clouds are smaller in area than a typical grid resolution of climate models. We present a statistical model for spatial and temporal distribution of cloud cover and attempt to link large scale variables such as relative humidity with cloud cover. The model uses a small number of Fourier coefficients of the radius vector function to describe the contour function of a cloud. Cloud cover data analyzed is three months of hourly infrared data obtained by a satellite from the TOGA COARE experiment. Results indicate that this model is reasonable for describing cloud cover distribution over time.
SPEAKER 2:
NAME OF PARTICIPANT AND AFFILIATION:
Z.Q. John Lu, National Center for Atmospheric Research
TITLE OF PAPER:
Statistical Design for Adaptive Weather Observations
CO-AUTHORS AND AFFILIATIONS:
L. Mark Berliner, National Center for Atmospheric Research and Ohio State University
Chris Synder, National Center for Atmospheric Research
ADDRESS:
Z.Q. Lu National Center for Atmospheric Research P. O. Box 3000 Boulder, CO 80307-3000
PH: 303-497-1702 FAX: 303-497-1333 zlu@ncar.ucar.edu
KEYWORDS:
Design for nonlinear models; sequential estimation; Bayesian experimental design; data assimilation.
ABSTRACT:
Motivated by the problem of how to collect adaptive weather observations in the Fronts and Atlantic Storm Track Experiment (FASTEX), this paper discusses the statistical issues involved in the experimental design. THe goal is to select optimal sites for improving data assimilation and weather prediction in high-dimensional nonlinear systems. A Bayesian experimental strategy is proposed to deal with the sequential nature of teh problem and the plan is implemented on a toy 40-variable atmospheric model due to Edward Lorenz. Comparison is made with other approaches that are being investigated by the meteorological community. Some open issues are discussed.
SPEAKER 3:
NAME OF PARTICIPANT AND AFFILIATION:
Wendy Meiring, National Center for Atmospheric Research
TITLE OF PAPER:
Statistical challenges in analyzing stratospheric ozone data
CO-AUTHORS AND AFFILIATIONS:
Gary Grunwald, National Center for Atmospheric Research
Richard H. Jones, National Center for Atmospheric Research and University of Colorado, Denver
MAILING ADDRESS
Wendy Meiring National Center for Atmospheric Research P. O. Box 3000 Boulder, CO 80307-3000
PH: 303-497-1375 FAX: 303-497-1333 meiring@ncar.ucar.edu
KEYWORDS
Space-Time, Environmental
ABSTRACT
The analysis of stratospheric ozone data presents many challenges to the statistician. Data are available from several sources, including satellites and ozonesondes (balloon-based measurements). Periods of data collection are short relative to many of the processes of interest. Measurements are often at irregular spatial and temporal increments, and on different spatial and temporal scales. There are strong seasonal patterns in the mean and variability of ozone levels. Many other factors, such as the quasi-biennial oscillation, the solar cycle, and aerosol and chlorine levels are important covariates whose effects may vary with stratospheric level and season. Spatial and temporal correlation are present. We will discuss and illustrate some of these issues and summarize our analyses.
SPEAKER 4:
NAME OF PARTICIPANT AND AFFILIATION:
J. Andrew Royle, National Center for Atmospheric Research
TITLE OF PAPER:
A Bayesian Approach to Cokringing
CO-AUTHORS AND AFFILIATIONS:
L. Mark Berliner, National Center for Atmospheric Research and Ohio State University
ADDRESS:
J. Andrew Royle National Center for Atmospheric Research P. O. Box 3000 Boulder, CO 80307-3000
PH: 303-497-1704 FAX: 303-497-1333 royle@ncar.ucar.edu
ABSTRACT:
\documentstyle[11pt,epsf]{article} \renewcommand{\baselinestretch}{1.5} \setlength{\textwidth}{5.5in} \setlength{\evensidemargin}{0.5in} \setlength{\oddsidemargin}{0.5in} \setlength{\textheight}{8in} \setlength{\topmargin}{0in} \pagestyle{plain} \title{A Bayesian Approach to Cokriging} \author{ {\bf J. Andrew Royle} and {\bf L. Mark Berliner} \\ Geophysical Statistics Project \\ National Center for Atmospheric Research} \date{} \begin{document} \maketitle \begin{abstract} Let $(U(x),V(x)): x \in R^{2}$, be a bivariate random field and suppose that we wish to make predictions at some location $x_{o}$. When $U$ and $V$ are ``crosscorrelated'', this is the classical ``cokriging'' problem. However, in cokriging one attempts to estimate parameters of both the marginal and joint dependence of $U$ and $V$. One major problem encountered in this traditional approach is specification of models for both the autocorrelation (within variable) and crosscorrelation (across variables) such that the resulting joint variance-covariance matrix is valid. In practice, models that guarantee validity are not generally useful, or worse, this assumption is ignored. One may avoid this modeling difficulty by posing the bivariate prediction problem in a hierarchical framework, specifying models for marginal and conditional distributions, but not the joint distribution. This approach has other advantages. For example, one may have better knowledge of conditional and marginal behavior of the component processes than of the joint behavior. Additionally, it is natural to formulate conditional models for this bivariate problem as is done in the traditional bivariate normal regression setting. \end{abstract}
\end{document}
SPEAKER 5:
NAME OF PARTICIPANT AND AFFILIATION:
Christopher K. Wikle, National Center for Atmospheric Research
TITLE OF PAPER:
Hierarchical Bayesian Space-Time Models for Atmospheric/Oceanographic Processes
CO-AUTHORS AND AFFILIATIONS:
L. Mark Berliner, National Center for Atmospheric Research and Ohio State University
ADDRESS:
Christopher Wikle National Center for Atmospheric Research P. O. Box 3000 Boulder, CO 80307-3000
PH: 303-497-1722 FAX: 303-497-1333 wikle@ncar.ucar.edu
KEYWORDS: Markov chain Monte Carlo; Markov random field; Regimes
ABSTRACT:
Atmospheric and Oceanographic data typically contain many different scales of spatial and temporal variability. Such data may show trends, oscillatory phenomena of different scales, and regime shifts, and consequently are often non-stationary in space and time. In addition, these data sets often involve many observation/prediction locations and very long data records. These factors limit the effectiveness of traditional space-time statistical methods for atmospheric data. As an alternative, we have developed a flexible multistage hierarchical space-time model. The first stage of the model specifies a measurement error model for the observational data; typically, the ``true'' variable of interest plus error. The second stage of the model allows for site-specific time series models for this variable. This stage includes large scale (e.g., seasonal) variability plus a spatial-temporal dynamic process for the ``anomalies''. Much of our interest is with this anomaly process. In the third stage, the parameters of these time series models are themselves given priors (Markov Random Fields). The Bayesian formulation is completed in the last two stages by specifying priors on parameters. Given the large size of our data sets, we implement the model in a Markov chain Monte Carlo framework. The model is demonstrated on a data set relevant to the atmospheric/oceanographic sciences.