AR1 for non-stationary correlated noise

[JanJereczek]

Indicator summary

Correlated noise can lead to bad transition recognition if standard CSD is applied (variance, lag1-autocorelation, AR1 parameter). Therefore use an adapted formula that addresses this. Works bad for small window sizes.

Reference

• Boettner and Boers (2021), Critical Slowing Down In Dynamical Systems Driven By Non-Stationary Correlated Noise

Also of interest in that context: work of Frank Kwasniok (in preparation?)

Codebase

Nope

Implementation plan

See eq (17-20) of Boettner and Boers (2021)