Multivariate signals of population collapse in a high-throughput ecological experiment

Abstract

Anticipating population declines is a crucial goal of conservation ecology. Recent conceptual work suggests that populations facing growing stressors should exhibit sequential shifts in behavior, morphology, and abundance before declining to extinction. However, the lack of high-resolution, multidimensional data has hindered empirical validation of this conceptual work. Using an autonomously monitored, high-throughput experimental system, we generated individual-based data on populations of the ciliate Paramecium caudatum forced to collapse due to increasingly stressful conditions. The gradual introduction of a pollutant elicited the predictable sequence of responses—declines in movement speed, followed by declines in body length, emergence of early warning signals of collapse, and finally, abundance declines. Conversely, a press disturbance generated by the introduction of predators did not induce this sequence. The time between the first detectable trait changes and population collapse depended on the statistical approach used, but the sequence remained consistent. Using general additive models, detectable behavioral signals in the polluted populations occurred one generation before abundance-based early warning signals were detectable, and two generations before abundance decline. We highlight that multivariate monitoring, particularly individual-based metrics, is crucial for forecasting population declines.