Causality and persistence in ecological systems: a nonparametric spectral granger causality approach

Area of Science:

• Ecology
• Complex Systems
• Time Series Analysis

Background:

• Ecological systems exhibit complex dynamics with direct and feedback relationships.
• Inferring causality from time-series data is challenging, especially with periodic external drivers masking endogenous dynamics.
• Traditional Granger causality (G-causality) in the time domain struggles with periodic forcing.

Purpose of the Study:

• To explore spectral extensions of Granger causality (G-causality) for analyzing ecological networks.
• To evaluate the performance of spectral G-causality and its conditional extension in quantifying causal interactions.
• To improve the ability to link oscillatory behaviors in ecological networks to underlying causal mechanisms.

Main Methods:

• Utilized nonparametric spectral extensions of Granger causality (G-causality) to the frequency domain.
• Applied Fourier transform-based statistics to distinguish endogenous interactions from external periodic drivers.
• Investigated both standard spectral G-causality and its conditional extension for multivariate systems.

Main Results:

• Spectral G-causality enables scale-by-scale decomposition of causality in ecological systems.
• Conditional G-causality demonstrated superior performance compared to standard G-causality.
• The methods were validated using both synthetic and real ecological time-series data.

Conclusions:

• Spectral G-causality is a powerful tool for dissecting causal relationships in ecological networks, especially under periodic forcing.
• Conditional G-causality effectively identifies causal links and their associated timescales in complex ecological systems.
• These frequency-domain methods enhance our understanding of ecological dynamics and network interactions.