Grantee Research Project Results
2000 Progress Report: Combining Environmental Data Using Hierarchical Bayesian Space-Time Models
EPA Grant Number: R826887Title: Combining Environmental Data Using Hierarchical Bayesian Space-Time Models
Investigators: Zimmerman, Dale , Cowles, Mary Kathryn
Institution: University of Iowa
EPA Project Officer: Hahn, Intaek
Project Period: January 1, 1999 through December 31, 2001
Project Period Covered by this Report: January 1, 2000 through December 31, 2001
Project Amount: $199,904
RFA: Environmental Statistics (1998) RFA Text | Recipients Lists
Research Category:
Objective:
The objectives of this project are to: (1) to develop Bayesian hierarchical models for time-series, spatial and spatiotemporal processes that combine data from disparate sources; (2) develop efficient Markov chain Monte Carlo algorithms for implementing such models, and to assess their convergence; (3) evaluate model performance using simulation studies; (4) demonstrate gains in efficiency and predictive ability using the combined data rather than any of its component parts individually; (5) fit models to one or more real data sets, each of which is a composite of data obtained from two (or more) different sources; and (6) make computer code for fitting such models publicly available.Progress Summary:
In this project year, we have moved from time-series and spatial models into a new hierarchical spatiotemporal model. At one level of the hierarchy, spatiotemporal correlations among values measured at point sites are captured by a separable structure involving a spherical spatial correlation function and an AR(1) model for the temporal component. At a higher level of the hierarchy, a conditional autoregressive prior is used to capture spatial correlations among region-specific random coefficients. We also have fit appropriate joint models that combine data on spatial or temporal processes with data on health effects of which the processes are predictors. Further specifics are given under specific aim 5.
Prior to joining the project team, Brian Smith, the new graduate research assistant, developed a library of remarkably fast C-language functions for carrying out linear algebra operations. He has begun converting our model-fitting programs to use these libraries. Preliminary benchmarks indicate that the programs will run about five times faster after conversion.
In work that Zimmerman is doing with David Holland of the EPA, substantial reductions in the kriging variance were obtained by combining spatial data from two networks. This work is ongoing but will be ready for submission for publication soon.
Cowles, Zimmerman, Christ, and McGinnis have fit the model described under Specific Aim 1 to annual data on snow-water equivalent measured at over 2000 sites in the western United States by four different methods during an 89-year period. Smith and Cowles fit a joint model to data from the Iowa Radon Lung Cancer Study in which they evaluated the association between residential radon exposure (with spatially-correlated measurements) and lung-cancer risk.
Cowles and Smith fit a model to data from an AIDS clinical trial in which repeated measurements of CD4 count and plasma levels of HIV RNA (short bivariate time-series data for each patients) were jointly modeled with time to clinical disease progression. An interesting feature of this model is that area-under-the-curve of log10- transformed RNA, rather than instantaneous RNA measurements, was the predictor of time to clinical progression. Although the application is medical rather than environmental, the modeling and computing techniques developed are applicable to problems in which environmental time series data predict another outcome.
Computer code for fitting the model described in Cowles, Zimmerman, Christ, and McGinnis is available for download from: www.stat.uiowa.edu/~kcowles.
An updated version of this code that is more flexible, faster, and more user-friendly will be posted in February 2001.
Code for fitting all the other models developed to data will be made available during the final project year.
Future Activities:
We will continue to work on demonstrating that gains in efficiency and predictive ability from a combined analysis can be substantial. Acid rain data from the CASTNet and NADP networks and snow water equivalent data will be the main areas of application. Also, we will continue our work on computational strategies for our MCMC samplers, with a goal to speed convergence and/or to handle larger spatio-temporal problems (i.e., more spatial locations, more time points). Finally, since the upcoming year is the final year of the project, we will spend time writing and revising manuscripts that report work already done.Journal Articles on this Report : 5 Displayed | Download in RIS Format
Other project views: | All 15 publications | 6 publications in selected types | All 6 journal articles |
---|
Type | Citation | ||
---|---|---|---|
|
Cowles MK, Zimmerman DL, Christ A, McGinnis DL. Combining snow water equivalent data from multiple sources to estimate spatio-temporal trends and compare measurement systems. Journal of Agricultural, Biological, and Environmental Statistics 2002, Volume: 7, Number: 4 (DEC), Page: 536-557. |
R826887 (2000) |
not available |
|
Cowles MK. A Bayesian geostatistical model for comparing environmental data. ournal of Agricultural, Biological, and Environmental Statistics volume 2002;7:236 |
R826887 (2000) |
Exit |
|
Cowles MK. MCMC sampler convergence rates for hierarchical normal linear models: A simulation approach. Statistics and Computing 2002;12(4):377-389. |
R826887 (1999) R826887 (2000) |
not available |
|
Cowles, MK. Efficient model-fitting and model-comparison for high-dimensional Bayesian geostatistical models. Journal of Statistical Planning and Inference 2003;112(1-2):221-239. |
R826887 (2000) |
not available |
|
Isaacson JD, Zimmerman DL. Combining temporally correlated environmental data from two measurement systems. Journal of Agricultural, Biological, and Environmental Statistics 2000;5(4):398-416. |
R826887 (1999) R826887 (2000) |
Exit |
Supplemental Keywords:
global climate, acid rain, monitoring, watersheds, indoor air, water, health effects, human health, carcinogen., RFA, Economic, Social, & Behavioral Science Research Program, Ecosystem Protection/Environmental Exposure & Risk, Ecosystem/Assessment/Indicators, Ecological Effects - Environmental Exposure & Risk, Environmental Statistics, risk assessment, hierarchical Bayesian space-time models, Bayesian space-time model, co-pollutant effects, global environmental data, prediction, simulation studies, streamflow, statistical models, combined data, data analysis, spatial-temporal methods, data models, global warming, innovative statistical modelsRelevant Websites:
http://www.stat.uiowa.edu/~kcowles
Progress and Final Reports:
Original AbstractThe perspectives, information and conclusions conveyed in research project abstracts, progress reports, final reports, journal abstracts and journal publications convey the viewpoints of the principal investigator and may not represent the views and policies of ORD and EPA. Conclusions drawn by the principal investigators have not been reviewed by the Agency.