COS 37-4 - An experimental test of predicted coextinction cascades in an Atlantic rocky intertidal food web

Abstract: Species are going locally extinct at an alarming rate, yet the ability to predict the consequences of these extinctions remains limited. One complication is the potential for coextinctions: when the local extinction of a prey species triggers additional, and potentially unexpected, declines in their consumers. These ‘cascading coextinctions’ are simulated using a computational tool in network science known as robustness analysis. Decades of theoretical and simulation-based work predicts that cascading extinctions will be common in nature, and can even trigger food web collapse. However, few studies have followed up the simulations to test whether the predicted coextinctions actually occur. By relying on robustness analysis of ecological networks, we could be systematically under-predicting or over-predicting the potential for food web collapse in nature.



In this study we ask: what is the relationship between the predicted and observed coextinctions? To answer this question, we compare predictions from robustness analysis against the results of a field extinction experiment. Using a systematic evidence synthesis, we assembled a meta-web of consumer-resource interactions for the Gulf of Maine rocky intertidal system. This meta-web contains 276 species and 3431 interactions, which are then subset using the species list from our field experiment. We simultaneously conducted a factorial press removal experiment on the intertidal coast of Allen Island, Maine. The experiment mimics the sequential loss of species in a robustness simulation using a space-for-time substitution, factorially removing the three most abundant basal resources (mussels, barnacles, and rockweed). Finally, we compare the identity, order, and number of species predicted to go extinct using robustness analysis of the food web against the observed experimental results.



Here we show a fundamental mismatch between network-based predictions and observed coextinctions. Experimental removal resulted in a cascade of local consumer declines with the removal of only one basal species. For example, when barnacles were removed, consumer abundance decreased 3-fold (β_control = 210.6 ± 1.0, β_removal = 65.3 ±1.1) and higher trophic levels went locally extinct. When run with a similar pattern of simulated basal species extinctions, robustness analysis dramatically under-predicted coextinctions. However, the pattern of losses following experimental barnacle extinction mainly propagated through non-trophic rather than trophic links, indicating the critical role that ecosystem engineers play in food web stability. Overall, experimental extinction generated unexpected and novel community states that were poorly predicted by the food web indicating the need to develop new theory that better explains the empirical realities of this ecosystem.