Using AI to better understand and model the earth system
International research team wins major grant to support work combining machine learning with physical models of atmosphere and land to improve climate modeling and methods
Oberpfaffenhofen, DE / Jena, DE / Valencia, Spain / New York, USA–October 11, 2019–An interdisciplinary team of researchers from the German Aerospace Center (DLR), the Max Planck Institute for Biogeochemistry, the University of Valencia, and Columbia Engineering has been awarded a 2019 European Research Council (ERC) Synergy Grant to understand and model the Earth system with machine learning, one of the important approaches of artificial intelligence (AI). The prestigious award–10 million euros over six years–will support the team’s groundbreaking work in rethinking the development and evaluation of Earth system models, which are the basis for understanding and projecting climate change.
Prof. Veronika Eyring from DLR’s Institute of Atmospheric Physics and corresponding Principal Investigator (cPI) says: “We teamed up to join forces and combine our multidisciplinary expertise in climate modeling, terrestrial ecosystems, machine learning, and cloud parametrizations to address some of the main limitations in the simulation and analysis of climate change. This will allow us to better understand processes and to discover unknown causes and drivers in the Earth system.”
The motivation for the newly funded ERC project “Understanding and Modeling the Earth System with Machine Learning” (USMILE) is that there are still some fundamental limitations in understanding the Earth system, which also limits our ability to accurately simulate climate change. While Earth system models have improved significantly in the past decades, the models’ ability to simulate both global and regional Earth system responses, which are key for assessing climate change and its effects on the planet’s ecosystems and populations, is limited by the representation of physical and biological small-scale processes, such as clouds, stomata, and microbes.
“Our central hypothesis is that this lack of understanding can be solved using machine learning. Firstly, we now have a massive amount of Earth observation data, with unprecedented spatial and temporal coverage for many processes. Secondly, high-resolution cloud-resolving models are now available that explicitly resolve small-scale processes such as clouds. But those simulations are computationally very expensive, and can therefore only be run for a short time,” says Prof. Pierre Gentine, co-PI of the project from Columbia University’s School of Engineering and Applied Science.
“And thirdly, the field of machine learning has quickly evolved, enabling breakthroughs in the detection and analysis of complex relationships and patterns in large multivariate datasets. We can now not only fit and model complex functions but also learn causal relations,” adds Prof. Gustau Camps-Valls, co-PI of the project from the University of Valencia.
The team will develop machine learning algorithms to enhance Earth observation datasets accounting for spatio-temporal covariations, as well as developing machine-learning-based parametrizations and sub-models for clouds and land-surface processes that have hindered progress in climate modeling for decades. In addition, they will detect and understand modes of climate variability and multivariate extremes, and uncover dynamic aspects of the Earth system with novel deep learning and causal discovery techniques.
Traditionally, physical modeling and machine learning have often been treated as two different worlds with opposite scientific paradigms: theory-driven versus data-driven. “Even though it has extraordinary potential, machine learning has not yet been widely adopted to address the urgent need for improved understanding and modelling of the Earth system. We hope that, by bridging between physics and machine learning, we will be able to revolutionize Earth system modeling and analysis, leading to more robust climate projections on the long-term,” says Prof. Markus Reichstein, co-PI from the Max Planck Institute for Biogeochemistry. He adds: “USMILE can drive a paradigm shift in the current modeling of the Earth system towards a new data-driven, physics-aware science.”
ERC Synergy Grants are awarded to groups of two to four co-PIs who have complementary skills, knowledge and resources, and can jointly address research problems that could lead to breakthroughs not possible by the individual PIs working alone. The four PIs on the USMILE project all work at the intersection of Earth system and data science with complementary expertise. “We are excited to work together on this interdisciplinary team and thank the ERC for giving us this great opportunity,” says Prof. Veronika Eyring.
Set up by the European Union in 2007, the European Research Council is the premier European funding organization for excellent frontier research. Every year it selects and funds the very best, creative researchers of any nationality and age to run projects based in Europe. In most cases, ERC Synergy groups are interdisciplinary, often using multidisciplinary approaches, and meet regularly over the course of the project.
German Aerospace Center
The German Aerospace Center (Deutsches Zentrum für Luft- und Raumfahrt, DLR) is the national aeronautics and space research center of the Federal Republic of Germany. Its extensive research and development work in aeronautics, space, energy, transport, security, and digitalization is integrated into national and international cooperative ventures. DLR is also responsible for the planning and implementation of Germany’s space activities on behalf of the federal government. DLR is also the umbrella organization for one of Germany’s largest project management agencies.
Max Planck Institute for Biogeochemistry
The Max Planck Institute for Biogeochemistry, Jena, is part of the German Max Planck Society. Its mission is to investigate biogeochemical cycles (carbon, water, nutrients) from local to global scales with approaches including manipulation experiments, long-term observations and data- and theory-driven modelling approaches.
University of Valencia
The University of Valencia is a research and educational institution more than 500 years old, and one of the top universities in physics in Spain. The Image Processing Lab (IPL) is a multidisciplinary institute that gathers more than 60 faculty and researchers in Earth observation, image processing, vision science and machine learning. The lab has active collaborations with ESA, NASA and EUMETSAT for designing new sensor devices, measurement campaigns, and data-intensive processing chains. The IPL faculty educates the new generation of physicists, data scientists, remote sensing analysts, mathematicians and electrical engineers in various (under)graduate, master and doctoral programs.
Columbia University’s School of Engineering and Applied Science
Columbia Engineering, based in New York City, is one of the top engineering schools in the U.S. and one of the oldest in the nation. Also known as The Fu Foundation School of Engineering and Applied Science, the School expands knowledge and advances technology through the pioneering research of its more than 220 faculty, while educating undergraduate and graduate students in a collaborative environment to become leaders informed by a firm foundation in engineering. The School’s faculty are at the center of the University’s cross-disciplinary research, contributing to the Data Science Institute, Earth Institute, Zuckerman Mind Brain Behavior Institute, Precision Medicine Initiative, and the Columbia Nano Initiative. Guided by its strategic vision, “Columbia Engineering for Humanity,” the School aims to translate ideas into innovations that foster a sustainable, healthy, secure, connected, and creative humanity.