AI-Powered 'Living Systematic Map' Revolutionizes Climate Policy Research

A wealth of evidence: Machine learning approach leads to a 'living systematic map' of climate policy research

Research on climate policy is growing exponentially. Of the approximately 85,000 individual studies ever published on policy instruments for mitigating global heating, a good quarter are from 2020 or later. A study led by the Potsdam Institute for Climate Impact Research (PIK) in the journal npj Climate Action, using machine learning methods, now shows how this vast knowledge is distributed -- by instrument, country, sector and policy level -- and identifies research gaps. A corresponding web tool, the "living systematic map," will help to guide science and policy. It will be continuously updated to reflect the current state of research. "Rather than directly providing answers to questions about the effects of climate policies, this study displays an overview of what has actually been scientifically studied so far," explains Max Callaghan, PIK researcher and lead author of the study. "On the one hand, this informs existing gaps and thus directions for primary research, including through funding. On the other hand, this overview facilitates evidence synthesis work, i.e. the summarization of the state of knowledge for governments, for example in the IPCC Assessment Reports." The study shows, among other things, that climate protection policies in the two countries with the largest greenhouse gas emissions -- China and the U.S. -- are the subject of particularly intensive research. By contrast, Africa still offers plenty of scope for new insights, with the lowest ratio of research work to enacted policies. The study also identifies a research gap for some smaller countries with particularly impressive emission reductions, namely Greece, Denmark and Iceland. An analysis by policy instrument shows that economic instruments -- and carbon pricing in particular -- attract significant research, but that there is a global research gap when it comes to regulatory instruments such as standards or bans. The study warns against "blind spots," for example, with regard to the complementary benefits of such instruments when used in combination with pricing instruments. There is also a research gap with respect to the industrial sector: it accounts for 23% of greenhouse gas emissions, 13% of implemented climate protection policies, but only 8% of the research. To cope with the enormous volume of individual studies, the research team used so-called machine learning models. These intelligent big data tools are first "trained" on a manageable number of texts using a learning algorithm, and then automatically look at crucial passages to extract the relevant information. These big data tools were applied to a query in the OpenAlex database, which yielded a good million potentially relevant studies, identifying the approximately 85,000 actually relevant studies and generating the map of research on climate policy. "With this study and the associated interactive web tool, we take a critical step towards enabling rapid and accurate responses to the climate crisis," says Jan Minx, also a PIK researcher and a co-author of the study. "Our research map is continuously updated and provides snapshot of the available evidence in real-time. It is the basis for an even more ambitious project: a Climate Solutions Evidence Bank, which would then summarize the existing knowledge on what climate policies work for decision-makers." Minx notes that thousands of climate policies have already been introduced, from carbon taxes to subsidies for electric cars. "We now need to answer the key question of what works in which context, and we need to do so in real time, with the help of artificial intelligence, automatically updated in light of new studies."

A wealth of evidence: 85,000 individual studies about climate policy

Research on climate policy is growing exponentially. Of the approximately 85,000 individual studies ever published on policy instruments for mitigating global heating, a good quarter are from 2020 or later. A study led by the Potsdam Institute for Climate Impact Research (PIK) in the journal npj Climate Action, using machine learning methods, now shows how this vast knowledge is distributed -- by instrument, country, sector and policy level -- and identifies research gaps. A corresponding web tool, the "living systematic map," will help to guide science and policy. It will be continuously updated to reflect the current state of research. "Rather than directly providing answers to questions about the effects of climate policies, this study displays an overview of what has actually been scientifically studied so far," explains Max Callaghan, PIK researcher and lead author of the study. "On the one hand, this informs existing gaps and thus directions for primary research, including through funding. On the other hand, this overview facilitates evidence synthesis work, i.e. the summarisation of the state of knowledge for governments, for example in the IPCC Assessment Reports." The study shows, among other things, that climate protection policies in the two countries with the largest greenhouse gas emissions -- China and the USA -- are the subject of particularly intensive research. By contrast, Africa still offers plenty of scope for new insights, with the lowest ratio of research work to enacted policies. The study also identifies a research gap for some smaller countries with particularly impressive emission reductions, namely Greece, Denmark and Iceland. An analysis by policy instrument shows that economic instruments -- and carbon pricing in particular -- attract significant research, but that there is a global research gap when it comes to regulatory instruments such as standards or bans. The study warns against "blind spots," for example with regard to the complementary benefits of such instruments when used in combination with pricing instruments. There is also a research gap with respect to the industrial sector: it accounts for 23 percent of greenhouse gas emissions, 13 percent of implemented climate protection policies, but only 8 percent of the research. To cope with the enormous volume of individual studies, the research team used so-called machine learning models. These intelligent big data tools are first "trained" on a manageable number of texts using a learning algorithm, and then automatically look at crucial passages to extract the relevant information. These big data tools were applied to a query in the OpenAlex database, which yielded a good million potentially relevant studies, identifying the approximately 85,000 actually relevant studies and generating the map of research on climate policy. "With this study and the associated interactive web tool, we take a critical step towards enabling rapid and accurate responses to the climate crisis," says Jan Minx, also a PIK researcher and a co-author of the study. "Our research map is continuously updated and provides snapshot of the available evidence in real-time. It is the basis for an even more ambitious project: a Climate Solutions Evidence Bank, which would then summarise the existing knowledge on what climate policies work for decision-makers." Minx notes that thousands of climate policies have already been introduced, from carbon taxes to subsidies for electric cars. "We now need to answer the key question of what works in which context, and we need to do so in real time, with the help of artificial intelligence, automatically updated in light of new studies."