|
Activity Number:
|
113
- Nonlinear and Nonstationary Dependent Processes: Modeling, Inference, and Applications
|
|
Type:
|
Invited
|
|
Date/Time:
|
Monday, August 9, 2021 : 1:30 PM to 3:20 PM
|
|
Sponsor:
|
Business and Economic Statistics Section
|
|
Abstract #316730
|
|
|
Title:
|
Locally Stationary Spatial Processes
|
|
Author(s):
|
Soumendra Lahiri* and Tucker Sprague McElroy and Daniel Census Weinberg
|
|
Companies:
|
Washington University and US Census Bureau and US Census Bureau
|
|
Keywords:
|
Non-Gaussian ;
Mixed increasing domain asymptotics;
Nonstationary spatial process
|
|
Abstract:
|
The stationarity assumption is often not appropriate for modeling spatial data over large spatial domains. While the observed spatial field may be amenable to modeling by stationary random fields over smaller parts of the domain, a framework is needed to allow for (smooth) variations both of the scaling (or the variance function) and of spatial interactions (i.e., the spatial dependence structure) across the entire spatial domain. In this paper, we give a formulation of locally stationary random fields that allows for such variations in the spatial covariance function. We present fairly general constructions in both the frequency and spatial domains, deriving an estimator for the local covariance based on irregularly spaced samples from the spatial process. We also establish consistency of the local covariance estimator and obtain an expansion for its mean squared error (MSE). Results from a moderate simulation study illustrate finite sample properties of the proposed estimator
|
Authors who are presenting talks have a * after their name.
