|
Activity Number:
|
586
- Frontiers of Multivariate Spatial Methodology
|
|
Type:
|
Invited
|
|
Date/Time:
|
Thursday, August 1, 2019 : 8:30 AM to 10:20 AM
|
|
Sponsor:
|
Section on Statistics and the Environment
|
|
Abstract #300516
|
|
|
Title:
|
Multivariate Analysis of High-Dimensional Non-Negative Responses Over Large Spatial Domains Using NNGPs
|
|
Author(s):
|
Daniel Taylor-Rodriguez* and Andrew Finley
|
|
Companies:
|
Portland State University and Michigan State University
|
|
Keywords:
|
Joint species distribution modeling;
dimension reduction;
Dirichlet process
|
|
Abstract:
|
Non-negative high-dimensional response vectors are pervasive in population ecology and environmental monitoring. Several metrics used to determine abundance for species in communities are recorded in this form. Modeling this type of responses is further complicated by the fact that they must usually account for a large number of species to represent communities across large spatial domains. Because at each location only a few species are present, these vectors are predominantly populated by zeros. With this in mind, we formulate a multivariate Gamma regression model with a latent Spatial Factor Nearest Gaussian Process representation that deals with both the high-dimensionality of the vectors and can cope with a large number of locations. Furthermore, the factor loadings matrices are assumed to be location specific, but are clustered using a stick-breaking processes prior to help retain the scalability in the number of locations. The methods are tested in simulations and are also applied to a dataset spanning a large region in interior Alaska.
|
Authors who are presenting talks have a * after their name.
