Selected intelligent methods

Referring to the title of the KPA, selected intelligent methods will be presented.

Intelligent methods are used both in the experimental acquisition of Earth system data and in the evaluation and analysis of these data, as well as for simulations.

Melts and fluids are involved in most geological processes in the Earth's crust and mantle. We are interested in how the atomic structure and dynamics influence their physical properties, such as density, viscosity, electrical conductivity or sound velocity. We also study the chemical speciation and hydration in aqueous fluids, which are important parameters for thermodynamic modeling. 

We perform molecular dynamics (MD) simulations using both classical and quantum mechanical models for the particle interactions. The latter so called ab initio molecular dynamics simulations are performed on supercomputers.

Realistic models are not only needed to interpret experimental data but also to predict properties of melts and fluids at extreme conditions of pressure and temperature that are not accessible experimentally. The construction of such models is a big challenge in (geo)materials research.....Read more

Yaping Shao, Andreas Hense, Konstantin Klein, Patrick Ludwig, Andreas Maier, Jürgen Richter, Isabell Schmidt, Christian Wegener, and Gerd-Christian Weniger, and Andreas Zimmermann

Studies of the DNA structure of present-day humans point to the genetic origins of Anatomically Modern Humans (AMHs) in Africa around 300 ka BP, followed by their dispersal over the Eurasian continent between about 60 and 40 ka BP and the Americas after 14 ka BP. We develop a Human System Model (HSM) to reconstruct the human dispersal history. The human system is extremely complex with a large degree of freedom and a wide range of non-linearly interactions subject to stochastic forces. Machine-learning techniques are used to estimate the human existence potential (HEP, Klein et al. 2021, 2023; Shao et al. 2021), a key human-system variable encompassing the various processes on the biological, cultural and environmental dimensions. To do this, we combine archaeological data and climate data from paleoclimate model simulations (Shao et al. 2021, Ludwig et al. 2018) using a logistic regression model (the HEP model). The HEP model is trained for N = 1000 times, each with a different randomly selected subset containing 80% of the archeological-site presence/absence records. The trained model is evaluated using the remaining 20% of the presence/absence records.

As an example, we investigate how climate influenced the dispersal of humans of the Aurignacian in Europe. The Aurignacian techno-complex developed in the middle of the Last Glacial Period (LGP, 115–11.7 ka BP), during which the climate conditions underwent major changes. Figure 1 shows the patterns of HEP for the Aurignacian for the interstadial (Fig.1a) and stadial (Fig. 1b) times. In warm climate conditions, most archeological sites are found in the regions with HEP values larger than 0.5. In several areas, the model predicted high HEP, but no archaeological sites are found. At about 50^oN, a band of archaeological sites is recognizable, which defines roughly the northern border of human existence. For stadial times (Fig. 1b), the HEP values are much reduced and are larger than 0.5 only in isolated hot-spot areas in sheltered areas of large topographies. While the climate conditions become more challenging for humans in central and southwestern Europe, human existence remains possible. The HEP difference between the interstadial and stadial times (interstadial–stadial) is shown in Fig. 1c. The machine-learning technique reveals that the Aurignacian expansion most likely took place during interstadial times, weakened or stopped during stadial times, but even during stadial times, the conditions in Europe are sufficient for humans to survive and to adapt to the cold climate conditions. Our example shows that the coupling of machine-learning and dynamic modelling provides a powerful tool for human-system research never before possible.

References

Klein, K., Wegener, C., Schmidt, I., Rostami, M., Ludwig, P., Ulbrich, S., Richter, J., Weniger, G.-C., & Shao, Y. (2020) Human existence potential in Europe during the Last Glacial Maximum. Quat. Int., 581–582, 7-27. https://doi.org/10.1016/j.quaint.2020.07.046

Klein, K., G.-C. Weniger, P. Ludwig, C. Stepanek, X. Zhang, C. Wegener, Y. Shao (2023) Assessing Climatic Impact on Transition from Neanderthal to Anatomically Modern Human Population on Iberian Peninsula: a Macroscopic Perspective. Sci. Bull., in print.

Ludwig, P., Shao, Y., Kehl, M., & Weniger, G.-C. (2018). . Glob. Planet. Change, 170, 34-47. https://doi.org/10.1016/j.gloplacha.2018.08.006

Shao, Y., Limberg, H., Klein, K., Wegener, C., Schmidt, I., Weniger, G.-C., Hense, A. & Rostami, M. (2021). Human-Existence Probability of the Aurignacian techno-complex under extreme climate conditions. Quart. Sci. Rev., 203, 1067995. https://doi.org/10.1016/j.quascirev.2021.106995

Renewable energy transitions and climate change must be the two most pressing words of the 21st century. As a responsible society, we all look forward to sustainable solutions in both areas. There is one more word that is buzzing around us, called Artificial intelligence (AI), and many speculations are going around. Being a part of Prof. Crewell's team has provided me with the opportunity to delve deeply into all three aforementioned areas and uncover significant potential for interdisciplinary research in the future.

From an energy perspective, cloud systems play a significant role in modulating the solar radiation budget and the availability of solar radiation on the ground. They thus are of high interest for renewable energy applications. From a climate change perspective, they influence the distribution of heat and moisture around the planet. Significant uncertainty exists about their physical characteristics and missing links across various spatial and temporal scales.

Satellites capture a field of clouds, and very often similar, looking individual clouds form neighbors of each other, giving rise to spatial patterns. But from a satellite and AI perspective, we may have many cloud system observations; still, they need to be labeled/segmented to train a conventional supervised neural network. Therefore, from an AI point of view, my research focuses on learning without labels and letting the neural network learn from scratch, understand the underlying representation of satellite observations, and focus on extracting meaningful features which can be used further for data-driven process understanding of regional climate and exploit its usage for solar energy power production.