Copula-based semiparametric models for spatiotemporal data.

Area of Science:

• Statistics
• Environmental Science
• Data Science

Background:

• Spatiotemporal data analysis presents computational challenges, particularly with high dimensionality and non-Gaussian distributions.
• Existing models often rely on restrictive parametric assumptions, limiting flexibility.

Purpose of the Study:

• To introduce a novel copula-based spatiotemporal model for flexible analysis of complex data.
• To propose a semiparametric estimator that simplifies computation by modeling marginal distributions and dependence separately.

Main Methods:

• A copula-based spatiotemporal model is developed, utilizing nonparametric estimation for marginal distributions.
• The approach separates the modeling of marginal distributions and spatiotemporal dependence for computational efficiency.
• Conditional quantiles are used to construct point and interval predictions.

Main Results:

• The proposed semiparametric estimator is computationally simple and flexible.
• The model demonstrates superior accuracy in point and interval predictions for skewed data compared to normality-based methods.
• Validated through simulation studies and real-world wind speed data analysis.

Conclusions:

• The copula-based spatiotemporal model offers a robust and flexible alternative for analyzing non-Gaussian spatiotemporal data.
• The nonparametric approach enhances predictive accuracy, especially for skewed distributions.
• The method provides a practical tool for forecasting in environmental and other complex systems.