Staff
Jan Blunk
FB 4: Dynamik des Ozeanbodens
FE Magmatische und hydrothermale Systeme
- Phone:
- +49 431 600 2490
- Email:
- jblunk(at)geomar.de
Address:
Wischhofstraße 1-3
24148 Kiel
Research Interests
- Thermodynamic Modelling
- Aqueous Geochemistry
- Chemical Speciation
- Activity Coefficients
- Scientific Machine Learning (SciML)
- Knowledge Integration
Understanding marine geochemical processes often requires thermodynamic models that predict the composition of fluids and the chemical reactions that occur under different environmental conditions. These models rely on thermodynamic parameters that are constrained by experimental measurements. While extensive thermodynamic data are available for many compounds and aqueous systems, experimental measurements become increasingly scarce at high temperatures and pressures. As a result, thermodynamic models often have limited ability to generalize to environments such as hydrothermal systems, where temperatures and pressures exceed the range covered by available experimental data.
Existing thermodynamic databases contain decades of experimental measurements, yet thermodynamic data remain sparse under the extreme conditions encountered in many natural systems. As part of the MarDATA doctoral program, Jan investigates whether machine-learning methods can leverage existing thermodynamic data to estimate thermodynamic parameters beyond the range covered by experimental observations. By complementing established thermodynamic models, such approaches may help extend their applicability to a broader range of environmental conditions.
Projects
- 2026 – 2029: AI-queous – AI-derived thermodynamic parameters for aqueous modelling, funded by MarDATA as part of the Helmholtz Information & Data Science Academy (HIDA)
Vita
Since 2026
Research Assistant, Marine Mineral Resources (MMR), GEOMAR Helmholtz Centre for Ocean Research Kiel, within the Helmholtz School for Marine Data Science (MarDATA)
2023 – 2026
Research Assistant, Computer Vision Group, Friedrich Schiller University Jena
2021 – 2023
M.Sc. in Computer Science, Friedrich Schiller University Jena
Master's Thesis: Steering Feature Usage During Neural Network Model Training
2019 – 2021
B.Sc. in Computer Science, Friedrich Schiller University Jena
Bachelor's Thesis: Object Tracking in Wildlife Identification
2018 – 2019
Undergraduate Studies in Computer Science, Kiel University
Student Supervision
Co-supervision of undergraduate and graduate research projects in Computer Science, including two B.Sc. theses and two M.Sc. theses.
Publications
2025
, Bodesheim, P. and Denzler, J.
(2025)
Adaptive Model Selection for Expanded Post Hoc Debiasing and Mitigating Varying Degrees of Spurious Correlations.
In: Computer Analysis of Images and Patterns (CAIP 2025).
, ed. by
Castrillón-Santana, M., Travieso-González, C. M., Deniz Suarez, O., Freire-Obregón, D., Hernández-Sosa, D., Lorenzo-Navarro, J. and Santana, O. J..
Lecture Notes in Computer Science, 15622
.
Springer, Cham, Switzerland, pp. 101-111, 11 pp.
ISBN 978-3-032-05059-5
DOI 10.1007/978-3-032-05060-1_9.
, Camps-Valls, G., Creutzig, F., Fearnley, C. J., Han, B., Kornhuber, K., Rahaman, N., Schölkopf, B., Tárraga, J. M., Vinuesa, R., Dall, K., Denzler, J., Frank, D., Martini, G., Nganga, N., Maddix, D. C. and Weldemariam, K.
(2025)
Early warning of complex climate risk with integrated artificial intelligence.
Nature Communications, 16
.
Art.Nr. 2564.
DOI 10.1038/s41467-025-57640-w.
, Penzel, N. and Denzler, J.
(2025)
CausalRivers - Scaling up benchmarking of causal discovery for real-world time-series.
[Paper]
In: 13. International Conference on Learning Representations (ICLR). , 24.-28.04.2025, Singapore .
2024
, Penzel, N., Bodesheim, P. and Denzler, J.
(2024)
Beyond Debiasing: Actively Steering Feature Selection via Loss Regularization.
In: 45. DAGM German Conference, DAGM GCPR 2023, Heidelberg, Germany, September 19–22, 2023, Proceedings.
, ed. by
Köthe, U. and Rother, C..
Lecture Notes in Computer Science, 14264
.
Springer, Cham, Switzerland, pp. 394-408, 15 pp.
ISBN 978-3-031-54604-4
DOI 10.1007/978-3-031-54605-1_26.
2022
, Körschens, M., Brust, C. A., Käding, C. and Denzler, J.
(2022)
Pre-trained models are not enough: active and lifelong learning is important for long-term visual monitoring of mammals in biodiversity research—Individual identification and attribute prediction with image features from deep neural networks and decoupled decision models applied to elephants and great apes.
Mammalian Biology, 102
(3).
pp. 875-897.
DOI 10.1007/s42991-022-00224-8.
