Artificial intelligence could help to predict how endangered species are

The number of fish known to be in danger of dying out may just be the tip of the iceberg.

Currently just 300 of the more than 13,000 known species of marine fish are officially threatened with extinction. However, as almost 40% of marine fish have not been assessed, the true number could be far higher.

Getting through this backlog is difficult, as it requires vast amounts of time, money and research to systematically assess each species. While there’s no getting around the full process in the end, a new study offers a temporary fix through machine learning.

By training algorithms to recognise patterns in the characteristics of endangered fish, the researchers can estimate where currently unassessed species would likely sit on the Red List. Dr Diego Vaz, a Senior Curator of Fishes at the Natural History Museum, says this could allow conservation action to start while the fish waits for a formal assessment.

“Nothing will ever substitute the specific data for these fish, but as we don’t have enough research, resources or people to provide it in the short term, this provides a useful stand-in,” says Diego, who was not involved in the research.

“Most of the data that these models use will ultimately come from museum collections and good taxonomy. Knowing even a small amount about a species allows us to link it to its relatives, and start to make inferences about its lifestyle. This provides a basis to start taking steps to protect it.”

The research is published in the journal PLOS Biology.

Accelerating the Red List

The IUCN Red List is the gold standard in conservation biology. It assess the threatened status of individual species by using information including their lifestyle, habitat and population size. The species are then assigned one of seven classifications, from Extinct down to Least Concern.

However, putting these listings together is difficult. As Dr Rupert Collins, another Senior Curator of Fishes at the Natural History Museum, explains, it can take years for researchers to gather enough information on a species to decide just how endangered it is.

“You need a lot of data to perform a full IUCN assessment, covering various aspects of a species’ life history and habitat,” Rupert says.

“This may be near impossible for some species, where there might only very small number of specimens known from museum collections.”

These unlisted species are considered Data Deficient or Not Evaluated (DDNE). This means there is either not enough information about these animals to accurately categorise them, or that they haven’t been assessed yet.

While DDNE species wait to be assessed, there’s a serious risk that their situation could deteriorate. It’s estimated that more than half of all Data Deficient species are already threatened with extinction, with a very real concern that this could increase.

This makes it important to find out a species’ extinction risk sooner rather than later, with scientists now looking for stopgaps.

Machine learning is one promising option. These computer models are trained to recognise how characteristics like size or habitat are related to a species’ extinction risk. In some well-known groups, like mammals, these algorithms can be up to 92% accurate in predicting the conservation status of previously assessed species.

The challenge is now to extend these algorithms to look at groups which are less well understood, like marine fishes. 

Finding threatened fishes

In total, almost 5,000 species of marine fish are currently DDNE. The researchers wanted to change this, and so trained two different types of algorithm and set them to work on categorising these unassessed species.

When the two algorithms agreed, a fish was assigned as either threatened or unthreatened. If they didn’t, then it was left as Data Deficient.

The team found that over 1,300 fish species currently considered DDNE were likely threatened, which would bring the proportion of marine fishes at risk of extinction up from the current 2.5% to 12.7%. These species are generally large, slow growing fish that live in restricted areas of the ocean.

But even after the work of the algorithms, around 1,000 marine fish were still considered DDNE. Rupert says that he’s “not surprised” how many species are still a mystery to conservationists.

“There are still so many groups of fishes that we just don’t know enough about, including whether they even exist in the first place,” he says. “There are thousands of fish species that are yet to be named, and so performing a conservation assessment on a fish without knowledge of how to even identify it is rather difficult.”

“It’s likely that a large number of these fish are going through a silent extinction, where we don’t even realise they’re gone until it’s too late.”

Protecting the unknown

To try and prevent these possible extinctions, the researchers also investigated the areas where the threatened fishes live to highlight the importance of these interim assessments.

In addition to biodiverse regions such as the South China Sea and the Philippine Sea, the algorithms also identified the oceans around the poles and the western coasts of Australia and North America as having many fishes in need of protective measures.

At the same time as protecting known species, scientists will need to redouble their efforts to identify the unknown fish. With thousands of potential species yet to be discovered, these animals could alter our understanding of these fish hotspots still further.

“It’s a tough job to try and protect something that you don’t know exists,” Rupert says. “Extending these algorithms to try and interpret where undiscovered species might need protections would be a very useful next step, and would give us a chance to find them before they become extinct.”