|
Abstract:
|
This roundtable focuses on the statistical challenges to model, predict, and estimate the complex spatial-temporal structure of environmental data. One of the main challenges is to calculate a Gaussian likelihood for large space-time problems, and modern approaches to this problem use spectral methods and algorithms to approximate the determinant and the inverse of a large matrix. However, the impact these approximations have on inference is not yet clear. Another issue is how to model a complex space-time dependency structure, and current work involves extensions of separable models and coregionalization models (in the multivariate case). These are popular models, but are they realistic? What is the best software to handle large spatial-temporal problems? What are the advantages/drawbacks of WinBugs, SAS, R, and all the R packages available to handle space-time data?
|
