|
Activity Number:
|
89
- Nonparametric Methods for Modern Data
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Monday, August 9, 2021 : 10:00 AM to 11:50 AM
|
|
Sponsor:
|
Section on Nonparametric Statistics
|
|
Abstract #317677
|
|
|
Title:
|
Temporal-Dependent Principal Component Analysis of Two-Dimensional Functional Data
|
|
Author(s):
|
Kejun He and Shirun Shen* and Lan Zhou
|
|
Companies:
|
Renmin University of China and Texas A&M University and Texas A&M University
|
|
Keywords:
|
Functional Principal Component Analysis;
Bivariate Splines;
Triangulation;
Kalman filter and smoother;
Global warming
|
|
Abstract:
|
In this paper, we propose a novel model to analyze the temporal-dependent two-dimensional functional data on an irregular domain. Illustrated by a study of Texas temperature, our method assumes that the functional principal component scores of two-dimensional functional data are temporally correlated and models the scores as latent time series. To overcome the challenge that the two-dimensional functions of interest are irregularly and sparsely observed, we use the bivariate spline basis on triangulations. All of these ideas are integrated into a unified model and an expectation-maximization (EM) algorithm along with Kalman filter and smoother is developed to estimate the unknown parameters. A simulation study is conducted to demonstrate that the proposed model outperforms its alternative. We finally use the proposed model to analyze the dataset of Texas temperature and the results are consistent with the scientific conclusions in domain knowledge.
|
Authors who are presenting talks have a * after their name.
