Shifting variance structure as an indicator of large-scale ecological change
Predicting population responses to large-scale change requires an understanding of how population dynamics vary over space and time. For a variety of important metrics, monitoring programs are generally assumed to provide such information. For example, observed indices may vary among repeated samples from a single location (e.g., a sampling site), from site-to-site within a habitat area (e.g., a lake), from area-to-area (e.g., lake-to-lake), and among years. Although variability has historically been viewed as an impediment to understanding population responses to ecological change, it can provide an important signal, rather than just being viewed as noise. We suggest that shifting variance structure can be indicative of population-level responses to large-scale influences. Mixed models allow for partitioning total variability of a response variable into component spatial and temporal parts. As such, we can explore if the structure of variation (i.e., variance components themselves), not just the total variance, is responsive to severe or large-scale perturbations. We analyzed long-term monitoring data for walleye (Sander vitreus) using negative binomial mixed models to explore the notion that an ecological perturbation may induce a shift in a population’s underlying variance structure.
In Oneida Lake, NY, temporal declines in gillnet catches of walleye were observed following the establishment of zebra mussels (Dreissena polymorpha). We predicted that decreased total catches would also be associated with a decrease in among-site variability, as a result of losing high abundance sites post-disturbance. As predicted, estimation-model results indicated that survey catches of walleye were more variable among sites historically than during recent years. Shifting variance structure between before and after perturbation periods could indicate that changes beyond reduced average catches over time have occurred across sites in the monitored population. Further, we expect that shifting variance structure could have implications for monitoring programs (e.g., influencing the power to detect long-term trends from standardized sampling). Improved understanding of how spatiotemporal variability responds to perturbation will be informative for forecasting future population dynamics in regards to anticipated changes in large-scale influences such as climate change.