There are so many studies on climate change and its consequences worldwide that human experts reach their limits when it comes to sifting through the results. Researchers have now trained an artificial intelligence to identify, evaluate and summarize scientific publications on climate change and its consequences. With the help of machine learning, based on more than 100,000 studies, they show that the effects of man-made climate change can already be proven on 80 percent of the land area. 85 percent of the world’s population are therefore affected.
Climate change is causing rising temperatures, more unevenly distributed precipitation, more frequent extreme weather events, melting glaciers and rising sea levels in the world. Countless studies deal with the various effects on the environment and people, partly on a local and regional level, partly globally. The IPCC evaluates many of these studies for its reports. But in view of the vast amount of scientific evidence, it is almost impossible for human experts to gain an overview.
Impact on the environment and society
A team led by Max Callaghan from the Mercator Research Institute on Global Commons and Climate Change in Berlin has now evaluated over 100,000 studies with the help of machine learning and created a world map on which the respective evidence for location-specific effects of climate change is recorded. “We conclude that the effects of man-made climate change can be found on 80 percent of the earth’s land area, where 85 percent of the population live,” the researchers write.
“These effects show up in a number of different systems and at different levels, spanning a broad spectrum of research areas, from glaciology to agronomy and from marine biology to migration and conflict research.” Previous overview studies have usually focused on one narrowed issue and were only able to include a limited number of studies on this. The approach of Callaghan and his colleagues, on the other hand, enabled a very extensive overview of the studies available so far.
Artificial intelligence helps with the evaluation
To do this, the researchers used an artificial intelligence that is able to evaluate natural language. “We further developed this software so that it can identify, localize and classify studies of observed climate impacts – on a scale that goes far beyond what is possible manually,” explain Callaghan and his colleagues. Using this technique, they found over 102,000 publications relevant to understanding the effects of climate change.
The artificial intelligence recorded, among other things, which region the respective results of the studies related to. The researchers compared this with further data on temperature and precipitation trends, which can be traced back to human activities with a high degree of probability. In this way, they created a world map that shows the regions for which the effects of man-made climate change have already been well documented. The result: “The majority of the world’s population lives in areas in which temperature and / or precipitation trends can be at least partially attributed to human influence,” according to the authors.
Data gaps in poorer countries
The researchers also found that the relevant publications are unevenly distributed across different regions. “Our results reveal a significant gap,” they write. “Robust evidence of potentially human-made climate change effects is twice as common for high-income countries as for low-income countries.” They found the strongest evidence for Western Europe, North America, and South and East Asia. In contrast, the evidence was lower for Africa and South America in particular. “This imbalance suggests that the lack of evidence in individual studies is due to the fact that these sites are being studied less intensively, rather than because there is no impact in these areas,” the researchers said.
According to the authors, their approach is well suited to summarize previous scientific evidence and clarify possible research gaps. “In general, it should be noted that this type of automated assessment process cannot replace careful assessment by experts,” they write. “However, the method can identify a large number of studies that can point to a human-made influence on the impacts of climate change.”
Source: Max Callaghan (Mercator Research Institute on Global Commons and Climate Change, Berlin) et al., Nature Climate Change, doi: 10.1038 / s41558-021-01168-6