How and why communities change is a central problem in community ecology. While there is debate over whether biodiversity patterns are changing directionally at a local scale, there is increasing evidence that many systems are experiencing turnover in species composition over time. Current methods of measuring temporal beta diversity often visualize change as linear, indicating an implicit assumption that community composition should change gradually in response to slowly changing drivers. While these approaches are effective for identifying directionality and magnitude of change, information on the temporal dynamics of community change can be lost. Recent work in ecological regime shifts has brought attention to the importance of rapid, non-linear response to slowly changing drivers. A machine-learning algorithm (Latent Dirichlet Allocation) capable of detecting slow, linear and fast, non-linear change in community species composition has recently been proposed for use in ecology. We apply this approach to a 38 year time series of a desert small mammal community that has previously been shown to be undergoing gradual long-term directional change using traditional approaches. We ask whether our interpretation of community dynamics is altered by our use of Latent Dirichlet Allocation.
Results/Conclusions
Over the 38 years of the study, the community experienced two reorganization events from one configuration to another. These reorganization events appear to be finite in length. While species composition changed dramatically (i.e. change in dominant species) across reorganization events, in-between reorganization events the community composition was relatively consistent. Our results show that the assumption of gradual, linear change in species composition is not well supported for this rodent community. Different mechanisms are expected to drive rapid and gradual community dynamics. Thus, quantifying the dynamics of community change, in addition to direction and magnitude, is an important consideration for understanding and predicting community assembly in the future.