The planet has entered the sixth mass extinction. Pollution, climate change and depleting resources could drive up to 27% of the world’s animal life to extinction, a new paper has claimed. The study used a supercomputer to map out how interdependent food chains could collapse in the coming decades.
Published on Friday and authored by European Commission scientist Giovanni Strona and Professor Corey Bradshaw of Flinders University in Australia, the study (Coextinctions dominate future vertebrate losses from climate and land use change, GIOVANNI STRONA HTTPS://ORCID.ORG/0000-0003-2294-4013 AND COREY J. A. BRADSHAW HTTPS://ORCID.ORG/0000-0002-5328-7741, DOI: 10.1126/sciadv.abn4345) presented a series of increasingly grim scenarios. The researchers claimed that the magnitude of the coming extinctions will depend largely on how much carbon mankind emits over the coming century.
A best-case scenario would see ecosystems lose 6% of their vertebrate species by 2050 and 13% by 2100, they claimed, while a worst-case outcome would see 10.8% of vertebrates wiped out by 2050 and 27% eliminated by 2100.
The scientists reached this figure by creating a “virtual earth,” populated by more than 15,000 “food webs” – complex relationships of plants, pollinators, predators, prey and parasites. With these species dependent on each other for food, the loss of one could lead to the collapse of an entire ecosystem.
A supercomputer scaled up these relationships to a global level, filled in missing data with fictional species, and simulated the effects of climate change on this virtual world. The computer also accounted for the impact of over-exploitation of resources, changes in land use and migration of species.
However, the researchers conceded that as it is based on a fictional version of our planet, their model “cannot forecast Earth’s future” with complete accuracy. Instead it describes a “realistic” future on an “ecologically plausible Earth.”
The study report said: Although theory identifies coextinctions as a main driver of biodiversity loss, their role at the planetary scale has yet to be estimated.
The scientists subjected a global model of interconnected terrestrial vertebrate food webs to future (2020–2100) climate and land-use changes. They predict a 17.6% (± 0.16% SE) average reduction of local vertebrate diversity globally by 2100, with coextinctions increasing the effect of primary extinctions by 184.2% (± 10.9% SE) on average under an intermediate emissions scenario. Communities will lose up to a half of ecological interactions, thus reducing trophic complexity, network connectance, and community resilience. The model reveals that the extreme toll of global change for vertebrate diversity might be of secondary importance compared to the damages to ecological network structure.
The study report said:
The planet has entered the sixth mass extinction. There are multiple causes underlying the rapid increase in observed and modeled extinction rates in recent times, of which land-use change, overharvesting, pollution, climate change, and biological invasions figure as dominant processes. However, assessing the relative importance and the realistic impact of such drivers at the global scale remains a challenge. Another aspect rendering assessment difficult are the synergies between drivers — a species might go extinct for multiple, simultaneous reasons, and in such contexts, ecological interactions play a fundamental role in predicting its fate. Growing recognition of the importance of species interactions in promoting the emergence of biodiversity in complex natural communities implies that an additional, fundamental component of biodiversity loss is represented by the amplification of primary extinctions across ecological networks. Coextinction — the loss of species caused by direct or indirect effects stemming from other extinctions — is now recognized as a major contributor to global biodiversity loss, strongly amplifying the effect of primary (e.g., climate-driven) extinctions.
It said:
Networks of ecological interactions are central to global patterns of diversity loss not only because coextinctions can be triggered by other extinction drivers, but also because network structure and dynamics might modulate several processes that can either reduce or increase extinction rate. For example, it is intuitive that a species’ success in colonizing a new area depends strongly on its ability to exploit local resources while simultaneously escaping enemies (predators and parasites). The addition of the new species might also initiate substantial changes to and have important cascading effects in the local network. Ignoring the structure of ecological networks and how they reconfigure as their constituent diversity changes therefore gives a possibly misleading view of the future of global diversity.
Previous attempts to predict the future of global diversity in the face of climate change and habitat modification have only considered the direct effects of these drivers on species (typically on single taxonomic groups), without explicitly accounting for ecological interactions.
The report said:
Apart from the obvious modeling and computational challenges to incorporate interactions among species, the main reason why there are few studies accounting for interactions is that obtaining sufficient data in most communities is intractable. Therefore, global-scale modeling of entire ecosystems appears to be the only viable solution, even if a challenging one. Recent developments in network approaches have shown that potential ecological interactions can be derived by applying different techniques (e.g., machine learning) to available datasets on species distribution and ecology. In previous work, we built on that idea to generate global-scale models of biodiversity by including species interactions using virtual species constructed to follow real-world archetypes. In such synthetic approaches, a virtual species is a plausible ecological entity that has a combination of ecological traits consistent with real-world species despite not corresponding exactly to them.
There are several advantages in using virtual species in this manner. The first is that once the rules have been set to generate virtual species, current gaps and biases in biodiversity sampling cease to be a limitation; we can use virtual species to populate the entire Earth and generate plausible ecological communities, even in areas where data on local diversity are scarce or missing. Second, virtual species avoid preconceptions (and biases) about current biodiversity patterns, permitting instead a focus on the processes involved in change. Here, we can populate an entire virtual planet with species, let them develop communities based on a modest set of realistic ecological rules and assumptions, and then explore the emerging patterns. With such an approach, real-world data serve as a template for generating the virtual species and for identifying the basic ecological rules controlling community dynamics and as a benchmark with which to validate the realism of modeled predictions.
In the models the scientists used, they found: In all climate scenarios, climate change was directly responsible for the most substantial fraction of local extinction events, followed by secondary extinctions, local extirpations due to overcompetition by colonizers, and land-use change.
The scientists could also quantify how the loss of species resulted in the loss of network interactions (edges). This loss of edges was substantial, averaging 23.6 ± 0.2% by 2050 in the worst-case scenario (47.0 ± 0.3% by 2100.
When the scientists compared the simulations including coextinction events to the controls that only accounted for primary extinctions, the average effects of coextinctions (measured as the percentage decrease in biodiversity between the coextinction and control simulations) were 27.5 ± 1.5%, 39.2 ± 2.5%, and 21.8 ± 0.6% by 2050 (27.1 ± 2.0%, 34.0 ± 4.0%, and 18.1 ± 0.7% by 2100) in the three climate-change scenarios SSP2-4.5, SSP4-6.0, and SSP5-8.5, respectively. However, a potentially overoptimistic assumption of the model is that herbivores and invertebrate feeders never run out of plant and insect biomass.
The study report said:
Our virtual-Earth model reveals the magnitude of and mechanisms driving biodiversity loss expected from climate change and land conversion this coming century. These results not only suggest a much greater loss than previously anticipated, they also demonstrate that biodiversity loss will be accompanied by an additional weakening of community resilience via erosion of the connectance of ecological networks.
The scientists said in the report:
While our virtual species are functionally realistic, they do not have taxonomic or phylogenetic meaning. Hence, our results reveal local changes in species diversity but do not provide information on global species extinctions per se. Neither does the model claim to produce an Earth replica, but instead aims to build an ecologically plausible Earth. Hence, the model cannot forecast Earth’s future but instead projects relative potential scenarios based on different assumptions (mainly carbon emissions) and reveals the underlying processes leading to those outcomes.
They said:
Examining the different drivers of extinction, our model reveals that the effect of climate change at the global scale is dominant, while land-use change played a comparatively minor role. However, in no way does that result refute the conclusion that land-use change is a major element of biodiversity loss; rather, it emphasizes that climate change is becoming more important. This emerges from two aspects of the model. First, we only considered relative land-use change from 2020 onward, meaning that the results reflect the relative future impact of land-use change, and not its dominant historical impact on biodiversity loss. Second, a strength of our model is that it can map extinctions everywhere on Earth. Even considering the extent of current human impacts, human presence and land-use change still directly affect only a small fraction of the total land inhabited by species. The area of primary and secondary cumulative land projected to be lost from 2020 to the end of the century for the worst-case climate change scenario (SSP5-8.5) is 8,000,000 km2 in total; however, this value represents only 6.5% of the global area in the simulation inhabited by at least one species in 2020 (~ 130,000,000 km2).
The scientists said:
Our approach reveals synergies among extinction drivers, thereby confirming that overexploitation of resources by novel colonizers combined with climate change (i.e., biological invasions) will become a major cause of diversity loss worldwide. Our model therefore provides the first global quantitative assessment of the impacts of biological invasions on planetary diversity over the coming century.
An interesting outcome from the sensitivity analyses is that increasing the frequency and intensity of acclimation events in local populations can counterintuitively lead to a higher global extinction rate. The “adaptation” mechanism implemented in our model assumes that species can shift their niches to match the climate of the preceding year. Such a mechanism — although reducing the risk of extinction for species capable of adapting under a stable climatic regime — does not necessarily ensure protection against future changes. When we assume a high probability of adaptation (e.g., 0.5% of species shift their niche in all localities every year), the net effect is a reduction in average persistence of global diversity. This suggests that a strong and widespread adaptation to local conditions might both increase robustness toward steady conditions while simultaneously increasing vulnerability to change. This outcome is consistent with predictions based on a completely different approach (artificial life evolution simulations) showing that as ecological networks become more resilient to stable environmental conditions, they also become increasingly susceptible to change.
The study report said:
The results confirm that coextinctions are fundamental drivers of mass extinctions and suggest that previous large extinction events revealed from the fossil record would likely have been exacerbated beyond their primary environmental drivers via the negative feedbacks arising from ecological dependencies. Unless conservation practitioners rapidly start to incorporate the complexity of ecological interactions and their role in extinction processes in their planning, averting the ongoing biodiversity crisis will become an unachievable target.
To incorporate as much plausible variation as possible in this stochastic model, the scientists generated an independent set of virtual species at the beginning of each global-extinction simulation according to the following procedure.
The scientists extracted species from the dataset associating species’ niches to the respective body mass, with the sampling structured to match known relative vertebrate diversity according to the IUCN (i.e., 5513 mammals, 7302 amphibians, 10,425 birds, and 10,038 reptiles).
The scientists then compared the body size of the selected species with all body sizes from the body-size and trophic datasets (in random order) until we found a species belonging to the same taxonomic grouping and with “matching” body size (see below), and then we associated the trophic information to the target species. In this way, we accounted for potential relationships between ecological niche, body mass, trophic level, and taxonomic group. At each step of the simulation, a species attempts dispersal from each locality to neighboring localities. Such a basic treatment of movement of species from one locality to another could conceivably increase diversity indefinitely. However, in the coextinction scenario where the scientists modeled food webs explicitly, the process is realistically constrained because species need to find their place in the food web that is conducive to survival. This is modeled so that when a new species enters a community based solely on climate compatibility, the local food web is rebuilt (according to the same rules described above), taking the new species into account. This might lead to different outcomes in each case.
For example, if the candidate colonizer is not a primary consumer and cannot find suitable resources, it cannot enter the food web, and colonization fails (i.e., it goes locally extinct). If a candidate colonizer does find a position in the food web, it can become associated with some resources and possibly become a resource for local consumers itself without driving those resources to extinction. Alternatively, if a candidate colonizer finds a position in the food web but ends up overexploiting one or more resources, it can lead to extinction of the resources and potentially itself and other species via extinction cascades.
However, if a candidate colonizer is a “primary” consumer (i.e., a vertebrate capable of consuming plants and/or invertebrates), it will always be able to find a position in the food web and become a resource for local consumers, producing an ecologically unrealistic accumulation of diversity and resources. This is because of the specific architecture of the model, where we assumed herbivores and insectivores are the “basal” components of the networks (i.e., the scientists assumed continued and unlimited availability of their resources). For this reason, they treated the incidences of insectivores and herbivores moving to a new locality as a particular case. Here, candidate herbivore and insectivore colonizers are added to the local community up to a maximum number of species, defined as the initial herbivore/insectivore diversity for that locality. If the diversity of insectivores and herbivores exceeds that number, the (herbivore and/or insectivore) species with least compatibility to local conditions are removed (i.e., considered as outcompeted by the other species).
In the scenario not accounting for coextinctions (i.e., not modeling food webs explicitly), the scientists applied the same criterion but extended to all species in the locality, regardless of their trophic ecology (with the only other criterion for colonization success being climatic compatibility). In this way, they treat the null expectation of primary extinctions only as a conservative case.
The scientists assumed that species have a certain capacity to adapt to local (and changing) climates unrelated to the traits defining phenotype.
The scientists ran 100 simulations per climate-projection scenario. In each simulation, they generated a pool of species, and populated a virtual Earth as described in previous sections. They did a burn-in phase where they permitted species to disperse to new areas over 100 time steps.
The scientists evaluated a species’ survival probability under the climate conditions from 2015 to 2020, and rebuilt food webs every 10 steps.
The scientists simulated climate and land-use changes from 2020 to 2100. At monthly steps and for each locality, they evaluated which species went extinct according to their suitability to withstand local temperature and precipitation conditions.
They avoided “random” extinctions, such that extinctions only occurred because of climate and/or land-use change.
They also assumed that in each year and in a given locality, a fraction of the species identical to the fraction of primary and secondary land lost from the previous year went extinct (see the next paragraph for additional details). Last, they simulated coextinctions in food webs due to primary extinctions and then updated the networks. They replicated the same simulations in a control scenario not accounting for networks and secondary extinctions.
The scientists opted for an intermediate scenario where the loss of species diversity is linearly proportional to land conversion. However, they also explored the potential effects on model outcomes of simulating different forms and magnitudes of response of local diversity to land use change.
