|
Activity Number:
|
131
- Methods for Spatial, Temporal, and Spatio-Temporal Data
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Monday, August 9, 2021 : 1:30 PM to 3:20 PM
|
|
Sponsor:
|
Section on Statistics and the Environment
|
|
Abstract #318748
|
|
|
Title:
|
A Semiparametric Approach for Prediction with Large Geostatistical Data Sets
|
|
Author(s):
|
Joshua French* and Mohammad Meysami
|
|
Companies:
|
University of Colorado Denver and University of Colorado Denver
|
|
Keywords:
|
semiparametric;
geostatistics;
big data;
nonparametric regression;
kriging
|
|
Abstract:
|
Environmental data sets with thousands to millions of spatially-referenced, correlated observations are becoming common. Modeling these data requires special care because (i) the data are frequently observed over a large area and have a complex, nonstationary mean structure and (ii) the dependence between observations must be modeled, which can be computationally challenging. Nonparametric regression is popular for estimating mean structure when a linear model is inadequate. Fixed Rank Kriging (FRK) models use a special parametric structure that allows for nonstationary spatial dependence while addressing computational challenges. We combine nonparametric regression with an FRK approach to modeling spatial dependence to produce a novel approach for modeling large geostatistical data sets observed over large areas. We compare prediction values and mean square prediction error for the proposed method and standard FRK using an application to real column-integrated CO2 data observed over the entire earth. Our results show that the proposed method can better model local features while simultaneously reducing the mean square prediction error.
|
Authors who are presenting talks have a * after their name.
