Climate Models Improved Through Artificial Intelligence and Earth Observation Data

Change in surface temperature. (Credit: © DLR/DKRZ)
  • At the UN climate conference COP26 in Glasgow, the world community advises on measures against climate change.
  • The impact of humans on climate change is clear, according to the World Climate Report.
  • As coordinating lead author, Prof. Veronika Eyring from DLR is jointly responsible for the report.
  • Her research evaluates different climate models with observational data from space travel and improves the models with AI.

COLOGNE, Germany (DLR PR) — The models for predicting climate change are becoming more and more accurate. They process huge amounts of data, evaluate information and combine them into an overall picture. The World Climate Report has shown what that looks like. “It is clear that human influence has warmed the atmosphere, the ocean and the land,” the report notes. The extent of the changes in the entire climate system is therefore unprecedented for many centuries to millennia. 

The fact that the statements are so clear is due to the work of Prof. Veronika Eyring. She is the coordinating lead author of the chapter “Human impact on the climate system”. The scientist from the German Aerospace Center (DLR) and the University of Bremen researches in the field of earth system modeling and model evaluation. It uses earth observation data from space travel and applies artificial intelligence (AI) methods to obtain reliable climate forecasts and technology impact assessments.

In Glasgow, Scotland, representatives from 197 countries are currently discussing measures to combat climate change: the climate conference COP 26 takes place from October 31 to November 12, 2021.

“Clear evidence of the urgency of action”

Droughts and heavy rain are increasing, ice in the seas is melting, the atmosphere is heating up: How can the more than 230 international experts who participated in the so-called Sixth IPCC Assessment Report have worked so be sure that human activities are causing climate change? 

“The lines of evidence have grown stronger over time. Not just for the temperature, but for many other climate changes, ”says Veronika Eyring. “We provided the reality check in the report. The warming has already risen to 1.1° C[elsius] compared to pre-industrial times. That means we are not far from the 1.5° C[elsius] target. The report also shows that every small increase in warming leads to wider and more severe effects of climate change. This is clear evidence of the urgency of action. The point now is to reduce greenhouse gas emissions immediately, quickly and drastically. Otherwise it will no longer be possible to limit the heating to a maximum of 1.5° C[elsius].”

ESMValTool analyzes how good the climate models are compared to earth observation data from space travel

In order to present the results of climate model simulations, the DLR Institute for Atmospheric Physics in Oberpfaffenhofen is in the lead together with more than 70 international research institutions Earth System Model Evaluation Tool (ESMValTool). The computer program allows a comprehensive evaluation of the climate and earth system models in comparison with observation data. The simulations of the latest generation of the so-called CMIP6 evaluated models and also addressed the problem of huge amounts of data. 

“We were able to show that the climate simulations have improved,” says Veronika Eyring, who headed the CMIP6 project from 2014 to 2020. In addition to observation data, the data products of CMIP6 represent an important source of climate information in the IPCC report.

Uncertainties and new research priorities in AI

Nonetheless, there are still uncertainties in predicting climate change and its effects on a global and regional scale. This is due in particular to the fact that small-scale processes such as cloud formation cannot be explicitly resolved and can therefore only be represented approximately in parameterizations. To solve this problem, Veronika Eyring relies on AI.

The interdisciplinary team at USMILE (Understanding and Modeling the Earth System with Machine Learning) develops machine learning methods to further improve the understanding and modeling of the Earth system. This is about a representation of processes in clouds and on the land surface. This is intended to reduce uncertainties in climate forecasts. 

The team is also working on examining climate fluctuations and extreme events, such as droughts, for causal relationships using methods such as deep learning. “Machine learning has extraordinary potential here for advancing climate research and opening up new fields of research,” says Veronika Eyring.

“We have the data and the infrastructure at DLR”

The comparison of earth observation data and earth system models is an important basis for improving climate predictions. “At DLR, we are predestined to develop these applications. We have the data and the infrastructure with which we can carry out the analyzes,” explains Prof. Dr. Markus Rapp, Director of the DLR Institute for Atmospheric Physics. 

The more extensive and precise the models, the more data is processed and checked. The team at the DLR Institute for Atmospheric Physics is creating new processes, expanding statistical methods and using artificial intelligence (AI) to further reduce the uncertainty in the forecasts and in the models. “AI helps us understand complex systems,” says Veronika Eyring.

Outstanding work in the field of climate modeling

Veronika Eyring supervises a dozen doctoral students in the field of climate modeling. “Many work on the interface between climate research and AI, that is, on the connection between physics and computer science. The young generation of scientists will still achieve a lot, ” says Veronika Eyring with certainty. 

She studied physics herself. When she was writing her diploma thesis in 1995, international climate researchers were already warning of man-made climate change. Veronika Eyring began to combine theoretical physics with climate research. In 2021 she received the renowned Gottfried Wilhelm Leibniz Prize for her outstanding work in the field of climate modeling because it has made a significant contribution to improving the understanding and accuracy of climate predictions through process-oriented modeling and model evaluation. 

The research prize money of 2.5 million euros [US $2.86 million] will be used for the further development of climate models through AI for improved technology assessment and policy advice – and for the training of young scientists. “The application of AI to understand and model the Earth system is still in its infancy. It is a promising area that requires a new generation of researchers trained at the interface between climate science and artificial intelligence, ”says Veronika Eyring.