A systematic approach for detecting abrupt shifts in ecological timeseries

Conservation efforts and sustainable use of natural populations often seek to reach or maintain viable abundance levels for a target population. Yet, this goal can be undermined by a number of events resulting from out-ofequilibrium dynamics, including large and sudden changes in abundance. The dynamical properties of such temporal changes are valuable indications about population's capacity to cope with environmental changes. Correctly identifying past or anticipating impending occurrences of temporal abrupt shifts in ecological systems is thus of major importance to adjust conservation and management strategies. Despite many available abrupt shift detection methods, few offer the possibility to compare and agree on the best model among linear, nonlinear, or abrupt models. By combining several existing methods, we develop an approach that classifies any timeseries to a trajectory type - no change, linear, nonlinear (quadratic), abrupt - and confirms the occurrence of potential abrupt shifts. We assessed the classification performances using a set of simulated data for which we had deterministic predictions for each type of trajectories. We used various levels of noise and perturbation events to make the simulations more realistic. This classification can be of particular interest when comparing dynamics of many populations across space or time. We show this by applying this classification approach to three different temporal datasets commonly used in conservation: catch tonnage, bird index, and insect occupancy timeseries. With this tool, we hope to promote conservation and management practices that explicitly take into account the likelihood of out-of-equilibrium trajectories and especially abrupt shifts in ecological systems.