Semiparametric estimation of cross-covariance functions for multivariate random fields

Area of Science:

• Spatial statistics
• Geostatistics
• Environmental modeling

Background:

• Joint modeling of multiple spatial processes is crucial for analyzing spatially referenced multivariate data.
• Existing methods often struggle with creating flexible yet non-negative definite multivariate spatial covariance functions.
• Accurate modeling of marginal and cross-process dependence is essential for spatial prediction.

Purpose of the Study:

• To propose a novel semiparametric approach for estimating multivariate spatial covariance functions.
• To enhance flexibility in modeling cross-covariance functions using spectral representations and B-splines.
• To improve spatial prediction accuracy compared to existing models.

Main Methods:

• Developed a semiparametric method utilizing spectral representations for flexible cross-covariance functions.
• Employed B-splines for specifying coherence functions to capture cross-spectral features.
• Utilized a likelihood-based estimation procedure and conducted simulation studies.

Main Results:

• The proposed method demonstrated strong performance in estimating coherence functions.
• Simulation studies confirmed the method's effectiveness.
• The approach outperformed the bivariate Matérn model and the linear model of coregionalization in spatial prediction tasks.

Conclusions:

• The semiparametric approach offers a flexible and effective way to model multivariate spatial covariance.
• This method provides superior spatial prediction for environmental data, such as particulate matter and wind speed.
• The technique advances the joint modeling of spatial processes, particularly in geostatistics and environmental science.