Machine learning and artificial intelligence to aid climate change research and preparedness

25 Nov 2019 Simon Davies

Machine learning and artificial intelligence are expected to provide significant new insights into understanding climate change and how to fight it, according to a study published today in Environmental Research Letters.

The study, led by the Centre for Ecology and Hydrology, UK, provides a compelling argument that machine learning (ML) and artificial intelligence (AI) can fill some of the gaps that exist in climate science.

Professor Chris Huntingford, from the Centre for Ecology and Hydrology, is the study’s lead author. He said: “Although climate research is generally considered to be a process-led activity, it is also extremely complex, and so the usage of statistically-based AI algorithms will elucidate new weather patterns and connections.”  

Co-author Professor Mike Bonsall said: “ML algorithms have advanced dramatically in recent years, and have enabled remarkable breakthroughs in other research sectors. There is every reason to expect they can be used to aid climate analysis.”

Co-author Hannah Christensen added: “There is a wide range of possible applications of ML in the climate sciences. While some people remain sceptical of ML, in fact climate scientists have been using tried-and-tested ML techniques for many years without realising it. We highlight this in our paper, as well as point out new areas where ML could transform our field.”

Co-author Dr Hui Yang commented: “Planet Earth is monitored at unprecedented levels, and especially by satellites collecting vast quantities of climate-related data. Researchers need more advanced algorithms and techniques to make better use of such huge amounts of data, in order to characterise trends, behaviours and interconnections.”

To illustrate specific potential of ML and AI, the researchers examined three examples where they could be used to gain more insight into climate events – the UK summer 2018 drought; the climate ‘hiatus’; and terrestrial ecosystem equation building – for example how plant nutrition interacts with climate and atmospheric chemistry and influences the amount of carbon that can be stored on the land surface.

Professor Huntingford said: “In terms of understanding the ‘hiatus’, it is essential that for a climate feature so prominent, we combine all strands of evidence to generate a definitive explanation. As the hiatus is likely a function of simultaneous interactions in the climate system, ML can aid characterise these, and point to any climate model deficiencies.

Co-author Dr Elizabeth Jeffers said: “In ecosystem equation building, ML can help identify which plant chemical traits and environmental conditions are driving feedback to the nitrogen and carbon cycles, enabling the development of process-based equations for use in modelling nutrient limitation.

Co-author Thomas Lees said: “Our study summarises where specific climate system components have already been investigated with ML. Our call is to go much further and employ ML methods to the entire the Earth system, with an emphasis on assessing its internal connections.

Professor Huntingford summarised: “Climate modelling needs a step-change to reduce uncertainty in projections – ML likely has a major role in achieving that.”