I propose a multiple time series model for data from a network of monitoring stations that have both temporal and spatial correlation. The model includes a separate mean and trend for each monitoring station and obtains spatial estimates of mean and trend by smoothing the observed values over a rectangular grid using a discrete smoothing prior. Smoothing parameters and covariance estimates can be chosen subjectively or selected using indirect generalized cross-validation. The gridded values and their standard errors can be used for several purposes, including inference on regional means or trends and improving monitoring networks via station rearrangement.
Bibliographical noteFunding Information:
* Gary W. Oehlert isAssociateProfessor,Department of AppliedStatistics, University of Minnesota, St. Paul, MN 55108. This research was supported in part by the U.S. Environmental Protection Agency (EPA) Atmospheric Research and Exposure Assessment Laboratory Contract #68D80063. This research has not been subjected to the review of the EPA and thus does not necessarilyreflectthe viewsof the agency, and no officialendorsement should be inferred. The author thanks Noel Cressie, Mike Stein, David Holland, Dennis Lettenmaier, the editor, and two anonymous referees for helpful comments and suggestions.
- Acid deposition
- Monitoring networks
- Smoothness priors
- Time series
- Trend analysis