B1841
Title: On partial envelop approach for modeling spatial-temporally dependent data
Authors: Wenbo Wu - University of Texas at San Antonio (United States) [presenting]
Abstract: Modeling multivariate spatial-temporally dependent data is a challenging task due to the dimensionality of the features and the complex spatial-temporal associations among the data across different locations and time points. To improve the estimation efficiency, a spatial-temporal partial envelop model is proposed which is parsimonious and effective in modeling high-dimensional spatial-temporal data. The partial envelop model is proposed under a linear co-regionalization model framework which allows heterogenous spatial-temporal covariance structure for different components of the response vector. The maximum likelihood estimator for the proposed model can be obtained through a Grassmann manifold optimization. A complete asymptotic result is obtained for the estimator and thorough empirical simulations are conducted to demonstrate the soundness and effectiveness of the proposed method. The proposed model is also applied to analyze the crowdsourcing weather data collected from personal weather stations in the United States.