Can we detect ecosystem critical transitions and signals of changing resilience from paleo‐ecological records?

Zofia E Taranu,Zoë Thomas,Marie‐Elodie Perga,Victor Frossard,Jean‐Philippe Jenny,Stephen R Carpenter,Jesse C Vermaire

doi:10.1002/ecs2.2438

Abstract

AbstractNonlinear responses to changing external pressures are increasingly studied in real‐world ecosystems. However, as many of the changes observed by ecologists extend beyond the monitoring record, the occurrence of critical transitions, where the system is pushed from one equilibrium state to another, remains difficult to detect. Paleo‐ecological records thus represent a unique opportunity to expand our temporal perspective to consider regime shifts and critical transitions, and whether such events are the exception rather than the rule. Yet, sediment core records can be affected by their own biases, such as sediment mixing or compression, with unknown consequences for the statistics commonly used to assess regime shifts, resilience, or critical transitions. To address this shortcoming, we developed a protocol to simulate paleolimnological records undergoing regime shifts or critical transitions to alternate states and tested, using both simulated and real core records, how mixing and compression affected our ability to detect past abrupt shifts. The smoothing that is built into paleolimnological data sets apparently interfered with the signal of rolling window indicators, especially autocorrelation. We thus turned to time‐varying autoregressions (online dynamic linear models, DLMs; and time‐varying autoregressive state‐space models, TVARSS) to evaluate the possibility of detecting regime shifts and critical transitions in simulated and real core records. For the real cores, we examined both varved (annually laminated sediments) and non‐varved cores, as the former have limited mixing issues. Our results show that state‐space models can be used to detect regime shifts and critical transitions in some paleolimnological data, especially when the signal‐to‐noise ratio is strong. However, if the records are noisy, the online DLM and TVARSS have limitations for detecting critical transitions in sediment records.

Highlights

The observation that ecosystems can respond discontinuously to changing external pressures has shed some light on their complex nonlinear dynamics
Observed lake sediment records To provide a range of regime shift time series, we cross-examined the published literature and identified core records with (1) an abrupt onset of a eutrophic state (Roxton Pond; echinenone pigment used as a proxy of total cyanobacteria abundance; Vermaire et al 2017); (2) a noisier eutrophication signal (Lake Anarry; myxoxanthophyll pigment used as a proxy of colonial cyanobacteria abundance; Stevenson et al 2016); and (3) possible flickering between alternate states (Lake Geneva; a composite proxy reconstructed from multiple intercorrelated varved cores tracking the inferred volume of hypolimnetic hypoxia; Jenny et al 2014)
When a critical transition occurred, our simulations showed that annualizing the record did not bias the resilience indicators and a discernible rise in rolling window variance was detected near the switchpoint (Appendix S1: Fig. S3c)

Summary

Introduction

The observation that ecosystems can respond discontinuously to changing external pressures has shed some light on their complex nonlinear dynamics. The concepts of critical transitions, thresholds, and alternative stable states spread throughout the ecological and environmental management literature, from the initial proposal by Holling (1973) to field evidence for ecosystem regime shifts (Scheffer et al 2001, Carpenter 2003). A regime shift, defined as a large change with prolonged consequences, is often obvious when it occurs, whether it qualifies as a catastrophic transition that pushes the system from one equilibrium state to another is usually difficult to prove (Carpenter 2003). A critical transition is defined as an unstable equilibrium point where the rate with which the system returns to this equilibrium (dominant eigenvalue) approaches zero, with the consequence that the system becomes increasingly slow in recovering from small perturbations (loss in resilience) and vulnerable to major changes caused by small perturbations. When approaching a critical transition, the system’s intrinsic rates will differ less from previous time points (increase memory or autocorrelation; Ives et al 2003) while its behavior becomes more like a random walk (rise in variance; Scheffer et al 2015)

Methods

Results

Discussion

Conclusion