You are here:
SPATIAL PREDICTION USING COMBINED SOURCES OF DATA
Citation:
MCMILLAN, N., D. M. HOLLAND, C. HOLLOMAN, AND G. YOUNG. SPATIAL PREDICTION USING COMBINED SOURCES OF DATA. Presented at EPA 2005 Science Forum, Washington, DC, May 16 - 18, 2005.
Impact/Purpose:
Our main objective is to assess the exposure of selected ecosystems to specific atmospheric stressors. More precisely, we will analyze and interpret environmental quality (primarily atmospheric) data to document observable changes in environmental stressors that may be associated with legislatively-mandated emissions reductions.
Description:
For improved environmental decision-making, it is important to develop new models for spatial prediction that accurately characterize important spatial and temporal patterns of air pollution. As the U .S. Environmental Protection Agency begins to use spatial prediction in the regulatory context, it will be increasingly important to combine output from atmospheric models with air monitoring data in a coherent way for improved spatial prediction, validation of model output, and for developing better linkages between air quality and public health outcomes. Typically air monitoring networks are sparsely, and irregularly spaced over large spatial domains, with monitors concentrated in urban areas. Output from numerical deterministic simulation models are produced over regular grids of size 36 km x 36 km or less, but generally have more bias in comparison to air monitoring data. By taking advantage of both types of spatial information, it is possible to provide improved maps of air pollution. We present a space-time hierarchical Bayesian modeling approach to predict daily fine particulate levels using Community Multi-scale Air Quality output and monitoring data from the EP A particulate monitoring network. An assessment of improved predictive performance using this method relative to a standard spatial prediction approach is made by predicting to sites of an independent network and calculating several goodness-of- fit statistics. This analysis is based on 2001 data in the northeast and midwest regions of the u.s.