Session Description: (short paragraph) The nonstationarity of physical processes is often caused by spatial or temporal effects. These effects may appear in either the trend, the dispersion, or the correlation structure. Very little work has been done in some of these areas, for example, in nonstationary correlations, even though real processes usually exhibit some of these complications. To prepare for a future of more complicated physical and engineering processes, we must do a better job of first modeling these processes so as to gain better understanding prior to performing inference.
The authors in this session will suggest strategies for analysis of such data from varied points of view. These works are motivated by real problems, and the methodologies will be illustrated using these data.
Theme Session: (yes or no) Yes
Applied Session: (yes or no) Yes
Session Chair & Affiliation: Jacqueline M. Hughes-Oliver, Department of Statistics, North Carolina State University
Mailing Address (include zip), Phone, Fax, E-mail: Department of Statistics, North Carolina State University, Box 8203, Raleigh, NC, 27695-8203 Phone 919 515 1954; Fax 919 515 7591; Email hughesol@stat.ncsu.edu
Session Organizer & Affiliation: Jacqueline M. Hughes-Oliver, Department of Statistics, North Carolina State University
Mailing Address (include zip), Phone, Fax, E-mail: Department of Statistics, North Carolina State University, Box 8203, Raleigh, NC, 27695-8203 Phone 919 515 1954; Fax 919 515 7591; Email hughesol@stat.ncsu.edu
Participant No. 1 & Affiliation: Paul Switzer, Department of Statistics & Department of Geological & Environmental Science, Stanford University
Co-Authors & Affiliations (for papers only): Sandra McBride, Department of Statistics, Stanford University
Title of Paper OR Role in Session: Time and Space Variation of Air Pollutants in an Indoor Environment
Key Words (for papers only): spatial variation, temporal variation, indoor air pollution
Abstract (for papers only): Recent experiments with tracer gases in an indoor environment exhibit complex spatial and temporal patterns. A source-proximity effect is evident but the magnitude of the effect is related to proximity in a complicated way. Short-term temporal variations are large close to the source and these affect the methods for elucidating the spatial structure.
Mailing Address (include zip), Phone, Fax, E-mail for Participant No. 1: Statistics Department, Stanford University, Stanford, CA 94305-4065 USA Phone: 1 415 723-2879 Fax: 1 415 725-8977 e-mail: ps@stat.stanford.edu
Participant No. 2 & Affiliation: Paul D. Sampson, Dept of Statistics, University of Washington
Co-Authors & Affiliations (for papers only): Peter Guttorp, Dept of Statistics, University of Washington Wendy Meiring, National Center for Atmospheric Research
Title of Paper OR Role in Session: Spatio-Temporal Modeling for an Hourly Air Quality Monitoring Network
Key Words (for papers only): Nonstationary spatial correlation; Space-time Cross-variogram; Ozone; Photochemical model.
Abstract (for papers only): We address the analysis and modeling of spatio-temporal field monitoring observations of tropospheric, mostly surface level ozone in the context of the assessment or ``operational evaluation'' of complex photochemical air quality models. We present a number of graphic demonstrations of the complex spatio-temporal covariance structure of hourly field observations for the San Joaquin Valley in California. Both the temporal and spatial covariance structure are heterogeneous or nonstationary: the (local) spatial covariance structure varies in both space and time while the temporal correlation manifests ``periodic autocorrelation'' that varies in space. We discuss a number of approaches to modeling this complex structure as a basis for computing optimal grid-cell estimates. We introduce an adaptation of the ``linear model of coregionalization'' from geostatistics for modeling the spatio-temporal cross-(hour)-correlations or cross-variogram. Spatial grid cell estimates are compared with predictions of the ``SARMAP'' Air Quality Model for the San Joaquin Valley.
Mailing Address (include zip), Phone, Fax, E-mail for Participant No. 2: Dept of Statistics, Box 354322 University of Washington Seattle, WA 98195-4322 ph: 206-685-2664 fax: 206-685-7419 email: pds@stat.washington.edu
Participant No. 3 & Affiliation: Timothy C. Haas, School of Business Administration, University of Wisconsin-Milwaukee
Co-Authors & Affiliations (for papers only): none
Title of Paper OR Role in Session: A Method for Statistically Assessing Spatio-Temporal Pollutant Trends and Meteorological Transport Models
Key Words (for papers only): local regression, kriging, transport modeling, Monte Carlo, hypothesis testing
Abstract (for papers only): Up to now, modeling and computational difficulties have impeded efforts toward a formal statistical assessment of large-scale spatio-temporal pollutant trends and the predictive performance of meteorological transport models. Until such statistical assessments can be made however, environmental regulators will not be able to always defend regulatory decisions, and physical process modelers will not be able to give convincing evidence of a proposed model's predictive validity. This talk gives a two-stage statistical method that allows such hypothesis testing and model assessment. Stage 1 is to estimate the global covariance matrix of the random disturbances of a pollutant deposition process and, using this global covariance matrix, Stage 2 is to conduct a Monte Carlo hypothesis test of either a pollutant trend hypothesis or a goodness-of-fit hypothesis of a meteorological transport model. This method is described and demonstrated using conterminous U.S. sulfate deposition data.
Mailing Address (include zip), Phone, Fax, E-mail for Participant No. 3: Timothy C. Haas, School of Business Administration University of Wisconsin-Milwaukee P. O. Box 742, Milwaukee, WI 53201
(414) 229-4360 (voice) (414) 229-6957 (FAX) haas@csd.uwm.edu
Participant No. 4 & Affiliation: George Christakos, Department of Environmental Sciences & Engineering, University of North Carolina at Chapel Hill
Co-Authors & Affiliations (for papers only): none
Title of Paper OR Role in Session: A Bayesian Maximum Entropy Analysis of Spatiotemporal Data
Key Words (for papers only): Bayes, maximum entropy, spatiotemporal data analysis
Abstract (for papers only): Most natural processes -such as pollutant concentrations, climatic parameters and vegetation variables- develop simultaneously in space and time. Observations and predictions of spatiotemporal natural processes are subject to considerable uncertainty. Therefore, any decisions based upon such observations and predictions are also subject to uncertainty. The quantitative assessment of spatiotemporal uncertainties and the modelling of statistical characteristics of natural processes are extremely important in a variety of applications concerned with pollution monitoring and control, exposure assessment, the study of hydrologic and climatic parameters, the characterization of fluid flows in porous domains, the management of natural resources, etc. Conventional methods of space/time analysis and mapping, like classical geostatistics, use only the available observations ("hard" data), and do not account for important prior information ("soft" data) like, e.g., interval-type data, higher-order moments, probability distribution-related data, geological knowledge and physical models. Previous attempts to incorporate prior information into the estimation process are mainly concerned with purely spatial natural processes, involve approximations and do not give definite rules for setting up prior probabilities from the available prior information. The Bayesian Maximum Entropy (BME) method maximizes both prior information and posterior probability subject to the measurements as well as prior information, constraints and past experience available. BME is a generalization of some traditional approaches; e.g., geostatistical kriging estimators and maximum loglikelhood can be derived as special cases of BME. The BME makes optimal use of spatiotemporal physical data, incorporates the various kinds of prior information available, provides a rigorous assessment of spatiotemporal variability and produces accurate predictive space/time maps. The method does not use any Gaussian-type hypotheses and offers a measure of the space/time mapping accuracy in each individual case rather than an average over all possible cases. Various features of the BME will be discussed and numerical applications will be given.
Mailing Address (include zip), Phone, Fax, E-mail for Participant No. 4: Department of Environmental Sciences and Engineering, University of North Carolina, CB#7400, 111 Rosenau Hall, Chapel Hill, N.C. 27599-7400 Tel.: (919) 966.1767 Fax: (919) 966.7911 E-mail: george_christakos@unc.edu
_________________________________________________________________
Jacqueline M. Hughes-Oliver 919-515-1954 Assistant Professor 919-515-1169 (fax) Department of Statistics hughesol@unity.ncsu.edu N.C. State University Raleigh, NC 27695-8203
_________________________________________________________________