Spatiotemporal Heterogeneity Learning: Generalized SpatioTemporal Semi-Varying Coefficient Models With Structure...

Area of Science:

• Statistics
• Environmental Science
• Data Science

Background:

• Spatiotemporal data analysis often struggles to differentiate between constant and varying effects.
• Existing models may lack the granularity to capture complex heterogeneity.
• Accurate modeling is crucial for understanding environmental factors influencing variables like particulate matter.

Purpose of the Study:

• To propose Generalized SpatioTemporal Semi-Varying Coefficient Models (GST-SVCMs) with structure identification.
• To enhance the detection and interpretation of spatiotemporal heterogeneity.
• To improve computational efficiency and statistical power in analyzing complex datasets.

Main Methods:

• Development of GST-SVCMs incorporating structure identification.
• Consistent estimation of constant coefficients and asymptotically normal estimation for statistical inference.
• Extension to Hierarchical SpatioTemporal Varying Coefficient Models (HSTVCMs) for refined structure identification.
• Monte Carlo simulations and real-world data application (particulate matter).

Main Results:

• GST-SVCMs accurately identify true model structures.
• The proposed method significantly improves prediction accuracy over models without structure identification.
• HSTVCMs offer more precise structure identification by decomposing effects into spatial, temporal, and spatiotemporal components.
• The approach provides insights into meteorological factors influencing particulate matter levels.

Conclusions:

• GST-SVCMs and HSTVCMs provide a robust framework for analyzing spatiotemporal heterogeneity.
• Structure identification enhances model interpretability, efficiency, and predictive power.
• The methodologies are effective for real-world environmental data analysis, informing policy and research.