Mechanism-Dependent Performance of Early Warning Signals for Critical Transitions: A Comparative Analysis

This study evaluates the performance of early warning signals (EWS) for critical transitions across mechanistically distinct collapse scenarios. While commonly used indicators such as lag-1 autocorrelation (AR1) and Fisher Information (FI) are often treated as universally applicable, their reliability under heterogeneous system dynamics remains insufficiently understood.   We conduct a comparative analysis of four indicators — AR1, Fisher Information, integrated information (Φ), and the Composite Resilience Transition Index (CRTI-2) — across three collapse mechanisms: (i) canonical saddle-node bifurcation, (ii) fold bifurcation with noise amplification, and (iii) structural compression collapse. Performance is evaluated using lead time, precision, recall, and F1 score under controlled stochastic simulations.   Results demonstrate that indicator performance is strongly mechanism-dependent. AR1 performs best in classical bifurcation scenarios consistent with critical slowing down. Fisher Information achieves higher precision under noise-amplified conditions, reflecting its sensitivity to distributional structure. In contrast, integrated information (Φ) provides a substantial lead-time advantage (1.4–1.9×) in structural compression regimes, where collapse is preceded by progressive decoupling of system components rather than scalar variance changes. CRTI-2 offers robust cross-mechanism performance but with reduced lead time compared to specialized indicators.   These findings do not support the hypothesis of a universal early warning signal. Instead, they motivate a mechanism-aware framework for indicator selection, formalized here as a 2×2 conceptual taxonomy linking collapse dynamics to indicator suitability. The results highlight the importance of integrating multivariate structural metrics alongside classical time-series indicators for reliable detection of impending transitions in complex systems.     early warning signals, critical transitions, regime shifts, Fisher Information, autocorrelation, integrated information, Phi, CRTI, complex systems, bifurcation theory, structural collapse, system resilience, tipping points, stochastic dynamics, complexity science