Many ecosystems are expected to abruptly change over the coming decades, yet predicting abrupt change is challenging because ecosystem state-driver relationships can take many forms. In systems with alternative states, a small driver change can lead to a large irreversible state change called a critical transition. Methods for anticipating critical transitions, called resilience indicators, employ expected statistical changes in system dynamics as the system approaches a critical threshold. Analysis of long ecological time series, as recorded by paleoecological proxies from sediment cores, are a promising avenue for testing resilience indicators and diagnosing causes of past abrupt regime shifts. However, data from sediment cores are subject to time-averaging, missing data, and uneven sampling. These processes are likely to affect resilience indicators, but their effects are poorly understood. Similar gaps between samples and time-averaging are common in many ecological data-sets.
Here we used a data simulation approach in which we built a model with alternative woodland-grassland states to simulate past intrinsic regime shifts characterized by critical transitions. We then explored the impacts of alternative sedimentation regimes and sampling designs on two widely-used resilience indicators—change in standard deviation and autocorrelation time—to assess how sedimentary processes transform resilience indicators and the ability of these approaches to distinguish past regime shifts from other ecological changes.
Results/Conclusions
The woodland-grassland model has alternative states and undergoes critical transitions resulting from gradual changes in carrying capacity, K (driven by precipitation). Abrupt shifts to grassland can also result from sudden changes in K, which do not generate resilience indicators.
The diagnostic ability of one resilience indicator, standard deviation, is retained under linear models of sediment accumulation rate and time averaging: critical transitions can be distinguished from extrinsic, non-critical, regime shifts when sedimentation rates are linear (uniform time averaging) or “broken stick,” characterized by a sharp break between two linear sedimentation regimes. Conversely, autocorrelation time is not robust to paleoecological transformations: critical transitions are not distinguishable from other regime shifts and other ecological changes under any sedimentation models examined here. Under exponential sedimentation, which is typical of upper columns of lake sediments still experiencing dewatering and compression, neither standard deviation nor autocorrelation time effectively distinguishes critical transitions from other ecological changes. Subsampling sometimes improved diagnostics and never had a negative effect. These results suggest that under non-exponential sedimentation, and with adjusted discriminant thresholds, standard deviation is a broadly effective metric for detecting critical transitions in paleoecological records.