Multivariate Modeling of Some Datasets in Continuous Space and Discrete Time

Area of Science:

• Environmental Science
• Geostatistics
• Time Series Analysis

Background:

• Multivariate space-time data are common in environmental science, often treated as time series.
• Existing geostatistical frameworks require practical covariance models for accurate characterization.
• Discrete, regularly monitored intervals necessitate specialized modeling approaches.

Purpose of the Study:

• To propose novel multivariate spatio-temporal covariance matrix functions.
• To model stochastic processes with autoregressive and moving average (ARMA) temporal margins.
• To ensure the validity and practical applicability of these covariance functions.

Main Methods:

• Deriving conditions for valid covariance matrices.
• Integrating time series analysis and spatial statistics methodologies.
• Applying proposed models to Kansas weather data using co-kriging.

Main Results:

• Development of new classes of multivariate spatio-temporal covariance functions.
• Demonstration of model validity and practical identifiability.
• Successful application to weather data prediction via co-kriging.

Conclusions:

• The proposed multivariate covariance functions offer a robust framework for spatio-temporal data.
• These models enhance prediction accuracy compared to traditional methods.
• The approach facilitates practical implementation in environmental geostatistics.