Quantifying changes in fish population stability using statistical early warnings of regime shifts

Jonathan A. Walter, Levi Lewis, James Hobbs, Andrew Rypel

Ecological conservation and management benefits from tools that can foresee impending problems, or those in early stages. Statistical early warnings of regime shifts, which can identify generic changes in system behavior associated with stability loss and potential abrupt changes to a new, distinct state, are theoretically well grounded and have been successfully applied in real-world settings. However, early warning indicators have seldom been applied to empirical animal population data. We quantified early warning metrics in 29 fishes using > 4 decades of monitoring data from the San Francisco Estuary and Sacramento-San Joaquin river system to develop an index describing the magnitude of evidence of population stability loss and potential regime shifts, relative to other studied species. Spatial synchrony increased in over twice as many species as it decreased, but temporal variance and lag-1 autocorrelation showed no tendency to have increased across species species. A composite early warning indicator (EWI) index developed from these metrics identified higher-risk species (e.g., white croaker, tule perch) from lower-risk ones (e.g., northern anchovy, fathead minnow). The composite index was uncorrelated with long-term abundance trends or whether the species is native or non-native. We also developed an index of confidence in the composite EWI score; considering both the EWI score and confidence index simultaneously suggests possible responses for research and management. For high EWI score, high confidence species may be candidates for targeted research and interventions, while high EWI score, low confidence species may be candidates for enhanced monitoring to better constrain population dynamics. Despite concerns about attributing changes in EWI metrics to regime shifts in short time series, there appears to be value in applying generic EWIs to population time series of animals with generation times ≥ 1 year, and approaches like ours may be valuable when little is known about organism life history, and when applying a consistent protocol can facilitate comparison across many species.