Interpretable, predictive spatio-temporal models via enhanced pairwise directions estimation.

Area of Science:

• Statistics
• Data Science
• Geospatial Analysis

Background:

• Predictive modeling for spatio-temporal data is crucial for understanding complex systems.
• Existing methods often rely on prespecified parametric models, limiting their ability to capture nonlinear structures.
• Interpreting spatio-temporal data requires methods that can effectively explore patterns and trends in both space and time.

Purpose of the Study:

• To develop a novel approach for predictive modeling and interpretation of spatio-temporal data.
• To capture nonlinear mean structures without relying on prespecified parametric models.
• To emphasize the exploration of geometric information within the data for enhanced understanding.

Main Methods:

• A supervised dimension reduction approach is developed.
• The Pairwise Directions Estimation (PDE) method is implemented to identify spatial patterns and temporal trends.
• The PDE method is enhanced to PDE+ by incorporating kriging for estimating unexplained random effects.

Main Results:

• The proposed PDE+ method demonstrates increased prediction accuracy.
• The method improves the interpretability of spatio-temporal models.
• Simulations and real-world applications show PDE+ is effective for exploring and interpreting spatio-temporal data patterns and trends.

Conclusions:

• The PDE+ method offers a powerful tool for analyzing spatio-temporal data.
• This approach effectively combines prediction with data exploration and interpretation.
• The study highlights the utility of geometric information in understanding complex spatio-temporal dynamics.