Extinction and range contraction have driven global food web collapse since Late Pleistocene

Since the Late Pleistocene, more than half of terrestrial mammal food web links have disappeared due to extinctions and range contractions, according to a new study, which uses deep-learning models to identify changes in global food webs over the past ~130,000 years. Not only do the findings underscore the impacts species loss has on the long-term persistence and function of ecosystems, but they also highlight the possibility of food web restoration through biodiversity recovery and conservation and provide a valuable framework for studying ecological networks in the past, present, and future. Human activities have driven contemporary extinctions and defaunation, which has caused cascading impacts on global biodiversity and ecosystem functioning through disruption of food webs. This is not only a modern phenomenon – widespread extinction and declines in species diversity since the last interglacial period nearly 130,000 years ago are also well known for many animal groups, particularly terrestrial mammals. However, these declines’ impact on food webs is challenging to understand as species-specific predator-prey interactions are rarely preserved in the fossil record. To better understand these dynamics, Evan Fricke and colleagues assembled a global database encompassing extinct and extant mammal traits, geographic ranges, and predator-prey interactions and used a machine learning approach to model changes in global terrestrial mammal food webs since the Late Pleistocene. Although only about 6% of terrestrial mammal species have gone extinct since the last interglacial, Fricke et al. found that more than half of global food web links have been lost, with particularly steep network declines following the first arrival of humans to a region and globalization events like European colonization. What’s more, while much of the global food web decline was born from extinctions that occurred long ago, range reductions for living mammals have resulted in a similar degree of decline, suggesting that restoration of the natural ranges of creatures could substantially restore food web complexity. “[Fricke et al.’s] deep-learning algorithm outperforms the allometric and phylogenetic models in common use by increasing the accuracy with which feeding interactions can be predicted,” writes Eoin O’Gorman in a related Perspective. “Although using machine-learning methods to study and model complex food web structures is not a new endeavor, it is yet to become a standard tool within the field.”