You are here:
A MOVING AVERAGE BAYESIAN MODEL FOR SPATIAL SURFACE AND COVERAGE PREDICTION FROM ENVIRONMENTAL POINT-SOURCE DATA
Citation:
Wolpert, R. L., E R. Smith, AND M. O'Connell. A MOVING AVERAGE BAYESIAN MODEL FOR SPATIAL SURFACE AND COVERAGE PREDICTION FROM ENVIRONMENTAL POINT-SOURCE DATA. Presented at US EPA 23rd Annual National Conference on Managing Environmental Quality Systems, Tampa, FL, April 13-16, 2004.
Impact/Purpose:
Provide regional-scale, spatially explicit information on the extent and distribution of both stressors and sensitive resources.
Develop and evaluate techniques to integrate information on exposure and effects so that relative risk can be assessed and management actions can be prioritized.
Predict consequences of potential environmental changes under alternative future scenarios.
Effectively communicate economic and quality of life trade-offs associated with alternative environmental policies.
Develop techniques to prioritize areas for ecological restoration.
Identify information gaps and recommend actions to improve monitoring and focus research.
There are two task objectives that reflect the work done by LCB in support of the ReVA Program objectives:
Provide information management, spatial analysis support, and data and information accessibility for the ReVA Program
Provide program management support, technology transfer, and outreach.
Description:
This paper addresses the general problem of estimating at arbitrary locations the value of an unobserved quantity that varies over space, such as ozone concentration in air or nitrate concentrations in surface groundwater, on the basis of approximate measurements of the quantity and perhaps of associated covariates at specificied locations. A nonparametric Bayesian approach is proposed, in which a joint prior distribution for the unobserved spatially-varying quantity is constructed as a moving average of independent-increment random measures. A reversible jump Markov chain Monte Carlo computational approach is proposed for approximating the posterior distribution of the unobserved quantity at all spatial locations, as well as averages of the quantity over arbitrary regions and other summaries of interest. The moving average Bayesian approach is compared with more conventional nitrate concentrations in groundwater. The surfaces and coverages are intended for use as part of the Regional Vulnerability Assessment (ReVA) program in the mid-Atlantic region.