Conditional heteroskedasticity warns of impending regime shift in whole-ecosystem experiment

Background/Question/Methods

Regime shifts are extensive rearrangements of nonlinear ecological processes that occur when ecosystems pass critical transition points. Some kinds of regime shifts, such as eutrophication, desertification, and loss of valuable species, may have important consequences for humans. Theoretical and laboratory evidence suggests that statistical anomalies, such as increased variance, may be detectable leading indicators of regime shifts in ecological time series, making it possible to foresee and potentially avert incipient regime shifts. There is a critical lack of field observations to test the efficacy of early warning signals at spatial and temporal scales relevant for ecosystem management. Conditional heteroskedasticity is persistent periods of high and low variance that may be a powerful leading indicator of regime shift. We evaluated conditional heteroskedasticity as an early warning indicator by applying moving-window conditional heteroskedasticity tests to time series of chlorophyll-a and fish catches derived from a whole-lake experiment designed to create a regime shift by manipulating the fish community. The purpose of our analysis was to evaluate the practicality of conditional heteroskedasticity as an early indicator using a known regime shift with high frequency data at scales relevant to ecosystem managers.

Results/Conclusions

Conditional heteroskedasticity was a powerful leading indicator that warned of the incipient regime shift about a year in advance. Conditional heteroskedasticity tests did not identify early warnings in the reference system. Early warnings appeared first in the fish catch time series. The response of chlorophyll-a was more delayed and was the consequence of slower evolving shifts propagating through the food web. We conclude that conditional heteroskedasticity is an effective leading indicator of regime shift in field settings. Probability values assess the chance of false positive results and provide a cut-off for objective interpretation by ecosystem managers.