Carl Zeiss Foundation supports interdisciplinary atmospheric physics and computer science project at Mainz University

Researchers expect that machine learning and methods of statistical data analysis will enable the evaluation of immense quantities of atmospheric physics data

27 July 2020

The Big Data in Atmospheric Physics (BINARY) project at Johannes Gutenberg University Mainz (JGU) has been granted EUR 1.5 million in funding by the Carl Zeiss Foundation. The basic idea of this interdisciplinary project that unites atmospheric physics and computer science is to apply state-of-the-art methods of statistical data analysis and machine learning to various scientific problems in atmospheric physics. Researchers from the Institute of Atmospheric Physics and the Institute of Computer Science at JGU are involved. The project was launched in March 2020 and is coordinated by Professor Peter Spichtinger from the Institute of Atmospheric Physics. "At JGU, we have excellent conditions for cooperation between the two disciplines of atmospheric physics and computer science. Over the past few years, we have increasingly carried out joint interdisciplinary research, also involving our students," said Spichtinger. The financial support of the Carl Zeiss Foundation is being provided under the auspices of the "Perspektiven 2018" research program and will continue for a period of five years.

Thanks to improvements in measurement technology and enormous increases in computing power, atmospheric physics – like many other natural sciences – has seen a huge increase in the amounts of data. However, analyzing these data using conventional methods has become difficult or even impossible. At the same time, our understanding of the intricate physical processes that occur in the atmosphere is still relatively poor as this is a highly complex system in which processes take place on many different scales – from the formation of microscopic ice crystals in the air to the development of thunderstorms. Indeed, many of the fundamental processes are still not fully comprehended while their effects on the system as a whole remain unclear.

Using state-of-the-art techniques of machine learning, particularly recently developed representation learning methods that employ deep networks, it has become possible to detect even more complex patterns in data and, if required, replicate them. Tools like these, which use statistical approaches to 'learn' models and structures from sample data, can open up new perspectives for the data analysis and modeling of complex processes – something that would be of particular benefit in the natural sciences and atmospheric physics in particular. Firstly, the extensive quantities of data in atmospheric physics could be analyzed, making it possible to identify relevant and dominant processes and to understand their effects. Machine learning methods would then facilitate the automatic propagation of models from this data. These could be used to find solutions to outstanding scientific problems in atmospheric physics.

Addressing scientific problems in atmospheric physics and computer science

The BINARY project brings together meteorologists and computer scientists of JGU to investigate relevant scientific issues in atmospheric physics using high-tech machine learning methods. As it is not usually possible and often not expedient to use standard methods without modifying these first, the researchers will be using an interdisciplinary strategy combining their expertise in computer science and atmospheric physics. The specific aspects of atmospheric physics they will be considering relate to the structural formation of clouds, the aggregation of clouds into larger systems, the improvement of the forecasting of challenging weather situations – such as the development of low stratus cloud – as well as how best to represent small-scale processes relating to coarse weather or climate models.

With these atmospheric physics objectives in mind, algorithms will be developed and/or adapted to suit the particular issue. This interdisciplinary concept allows the enormous potential of current machine learning techniques to be employed to serve the needs of a natural science discipline. The close cooperation is also expected to lead to advances in computer science, as it is likely that comprehensive algorithms will be developed from the complex application problems. The technical issues surrounding the handling of huge data sets are also problematic, particularly regarding storage and intelligent processing using the latest high-performance computing architectures. They also require the further development or adaptation of existing strategies.

However, it is not only atmospheric physics and computer science research that will gain as a result – the project is also intended to contribute generally to the digitalization of research methods at JGU. In this connection, there are plans to subject newly developed process modeling and pattern recognition tools to testing in practice in other disciplines, such as radiology and particle physics, and to make the related methods and insights available to other JGU researchers.

Favorable conditions for interdisciplinary research at JGU

The Institute of Atmospheric Physics at JGU has been offering lectures on the modeling of physical phenomena designed for computer science students for some time now, while collaborating with the Institute of Computer Science with regard to subjects for Bachelor's, Master's, and PhD projects. The BINARY project extends this cooperation to make major progress in this regard. Involved in the project are eight doctoral candidates from both disciplines who will work together, supported and supervised by ten experienced researchers. Doctoral candidates are still currently being sought for the natural sciences component. For more information, visit the project's website. It is also possible to work on topics for graduation theses (Bachelor's or Master's) while participating in the project.

Johannes Gutenberg University Mainz is supporting the project by making resources, such as the MOGON II mainframe, available as well as supplying the means to store the vast amounts of data. The project will also be providing input to the Mainz Institute of Multiscale Modeling (M3ODEL) core research area at JGU and will be supplemented by the work of the JGU Research Center for Algorithmic Emergent Intelligence, which is also funded by the Carl Zeiss Foundation. Thus, research being undertaken into quantitative multiscale modeling at JGU will also benefit from the BINARY project.

About the Carl Zeiss Foundation

The Carl Zeiss Foundation's mission is to create an open environment for scientific breakthroughs. As a partner of excellence in science, it supports basic research as well as applied sciences in the STEM subject areas (Science, technology, engineering, and mathematics). Founded in 1889 by physicist and mathematician Ernst Abbe, the Carl Zeiss Foundation is one of the oldest and biggest private science funding institutions in Germany. It is the sole owner of Carl Zeiss AG and SCHOTT AG. Its projects are financed from the dividend distributions of the two foundation companies.