Causal inference from cross-sectional earth system data with geographical convergent cross mapping

Area of Science:

• Earth System Science
• Complex Systems Analysis
• Spatial Data Science

Background:

• Temporal causation models face limitations with non-time-series or low-variation data, common in Earth systems.
• Existing spatial causation models inadequately explore rich spatial cross-sectional data.
• The generalized embedding theorem suggests causal links between variables within the same dynamic system.

Purpose of the Study:

• To propose a novel Geographical Convergent Cross Mapping (GCCM) model for spatial causal inference.
• To address the limitations of temporal models in Earth system science by utilizing spatial cross-sectional data.
• To develop a method for detecting and characterizing spatial causal relationships in complex systems.

Main Methods:

• Reconstruction of dynamic system state space from observational data.
• Application of cross-mapping prediction within the reconstructed state space.
• Development and testing of the Geographical Convergent Cross Mapping (GCCM) model.

Main Results:

• GCCM successfully detects weak-to-moderate causations even with insignificant correlations.
• The model identifies primary causation direction in strongly coupled variables.
• GCCM overcomes the mirroring effect, revealing bidirectional asymmetric causation.

Conclusions:

• The Geographical Convergent Cross Mapping (GCCM) model offers a robust approach to spatial causal inference.
• GCCM enhances the analysis of complex systems by leveraging spatial cross-sectional data.
• This method provides new insights into causality in Earth systems where temporal data is limited.