Master thesis: Machine Learning for High-Resolution Paleoclimate Modeling

Area of research

Diploma & Master Thesis

Job description

Master thesis: Machine Learning for High-Resolution Paleoclimate Modeling

Background: Coarse-resolution (~1°) global climate models are widely used in paleoclimate research but are unable to explicitly represent many small-scale oceanic and atmospheric features that shape regional variability and extremes. High-resolution climate simulations can resolve these processes but remain computationally expensive, strongly limiting their applicability for long simulations and ensemble studies. Recent advances in machine learning offer a promising alternative by enabling data-driven reconstruction of fine‑scale climate information from coarse model output (Oyama, 2023; Mardani, 2025).

Model performance will be evaluated using climate-aware diagnostics, going beyond traditional image similarity metrics. This includes assessing spatial variance, spectral characteristics, and the representation of extremes and coherent structures relevant for paleoclimate interpretation, such as marine heatwaves (Hayashida, 2020) or monsoonal precipitation (Liu et al., 2023). By combining modern machine learning techniques with state‑of‑the‑art climate simulations, your thesis will contribute to an emerging research direction at the interface of climate dynamics, paleoclimate modeling, and data science.

The main supervisor is Prof. Dr. Gerrit Lohmann. During the project you will be supported by two PhD students, one with a background in physics and the other with a background in data science, and be a part of the Paleoclimate Dynamics section at the Alfred Wegener Institute (AWI). We are an international and dynamic team:

Your Tasks

• You will explore machine learning–based methods for super‑resolving – translating low‑resolution into high‑resolution fields – paleoclimate simulations.
• Using paired climate model simulations of a past warm climate state (mid‑Holocene), you will train deep learning models – such as convolutional neural networks or diffusion models – to learn the statistical relationship between coarse and fine spatial scales.
• The focus will be on physically relevant variables such as sea surface temperature or near‑surface winds.

Your Profile

• Enrolled at a University
• Background in Physics, Oceanography/Meteorology, Applied Mathematics, or Data Science
• Experience with Python

  Further Information

  You are interested? Then please send us your application with cover letter and CV (merged into one PDF file) to Nina Öhlckers ( ) or Alexander Thorneloe ( ).

  This research center is part of the Helmholtz Association of German Research Centers. With more than 42,000 employees and an annual budget of over € 5 billion, the Helmholtz Association is Germany's largest scientific organisation.