Sublinear Algorithms for Meteorology

• Applicants: Professorin Dr. Susanne Crewell; Professor Dr. Christian Sohler
• Subject Area: Theoretical Computer Science ; Atmospheric Science
• Term: since 2025
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 559931366

Climate change is one of the major threats to mankind, and highly reliable monitoring and prediction of climate development is of the utmost importance to our society. One of the most relevant sources of uncertainty in climate prediction is our lack of understanding of how clouds react to climate change. Better resolving clouds in time and space in both satellite observations and climate models is seen as an important avenue to move forward. A main obstacle in this direction is the large amount of data that needs to be processed and that is still rapidly increasing. Processing such amounts of data requires a huge amount of computational resources using traditional methods which is not deemed possible. The limitations coming from this big data and high-compute setting only allow a handful of centers around the world to access the wealth of information, excluding a multitude of researchers and whole regions - not to mention the high amount of energy required that counteracts the fight against climate change. Such a scenario is clearly undesirable, and hence, it is important to find alternative ways of advancing research on climate change. In this project we propose a paradigm shift. Instead of trying to cope with larger and larger datasets by spending more and more computing resources, we aim to develop new specialized and highly efficient algorithms that build an intelligent summary of the data that (a) is significantly smaller than the original data set, (b) can be used for later data analysis tasks without a significant loss in accuracy, and (c) satisfies mathematically provable quality guarantees. In order to develop our new algorithms, we will build upon recent findings in the area of Sublinear Algorithms, a subdiscipline of Theoretical Computer Science, that aims to develop and analyze extremely efficient algorithms in big data settings. We will particularly rely on findings on so-called core-sets and sketches. These techniques are key to many data reduction problems related to data analysis. However, while core-sets and sketches are very effective data reduction techniques, they are also very problem dependent. Thus, the key challenge of the project will be to develop suitable core-set and sketch-based techniques for the data analysis problems at hand. We will show the viability of our approach by studying spectral radiance data from satellite measurements. Our goal is to analyze the data summaries using only a fraction of the resources required with traditional approaches while at the same time obtaining new insights into cloud characteristics and their governing processes supporting better weather and climate prediction. While primarily focusing on satellite data, we will consider novel hectometer climate model output potentially enabling a wide range of applications that can be explored in our strong regional collaboration network.

• DFG Programme: Reinhart Koselleck Projects