Machine Learning for Hydrological and Earth Systems
Dr. Shijie Jiang
Mission
In the face of increasing environmental and climatic pressures, it is imperative to improve our understanding of the variability and causality underlying biogeochemical cycles across different spatial and temporal scales. The research of the Machine Learning for Hydrological and Earth Systems (ML4HES) group, which is also part of the ELLIS Unit Jena, is at its core motivated by the objective of exploring how environmental and climate sciences can benefit from advances in machine learning and artificial intelligence. We focus on integrating data and domain knowledge with hybrid and explainable machine learning methods to improve the fundamental understanding of the interactions of climate, water, and ecosystems.
Focus areas
Hybrid and explainable AI in Earth system science
The main methodological focus of the group is knowledge-integrated and interpretable machine learning. We emphasize the combination of advanced data-driven techniques with classical Earth system analysis to improve our modeling and understanding of the complex dynamics among interconnected Earth subsystems. We view machine learning as both a bridge and a lens, connecting extensive Earth datasets to the underlying system dynamics. We are particularly interested in exploring effective and robust methods for identifying dependencies, interactions, couplings, feedback mechanisms, and causal relationships within various biogeochemical processes.
Key publications
- Jiang et al. (2024). How interpretable machine learning can benefit process understanding in the geosciences. Earth's Future, 12(7), e2024EF004540.
- Song et al. (2024). Towards data-driven discovery of governing equations in geosciences. Communications Earth & Environment, 5(1), 589.
- Jiang et al. (2020). Improving AI system awareness of geoscience knowledge: Symbiotic integration of physical approaches and deep learning. Geophysical Research Letters, 47(13), e2020GL088229.
Terrestrial ecohydrological processes and feedbacks
One of the main research goals of our group is to understand the complex interactions between vegetation and hydroclimate at multiple scales. Using advanced data-driven techniques and mechanistic ecohydrological and ecosystem models, we aim to uncover the intricate responses and feedbacks between water, plant physiology, and climate variables. We seek to understand how shifts in hydrological patterns can regulate ecosystem responses to environmental change, as well as how changes in terrestrial ecosystems can feedback into hydroclimate systems and influence extreme events.
Key publications
Blougouras et al.; Huang et al. (coming soon)
Water and carbon cycles in watershed systems
Another main focus of our group is to understand the processes that control water and carbon fluxes and ecosystem functioning at the watershed scale. We emphasize the role of watersheds as fundamental natural landscape units for the cycling of water, energy, carbon, and nutrients. Using advanced modeling techniques and data-driven analyses, we aim to uncover critical thresholds and interactions that influence watershed resilience to environmental change. Our overall goal is to identify, quantify, and predict hydrological and biogeochemical processes in response to natural and anthropogenic variability.
Key publications
- Jiang et al. (2024). Compounding effects in flood drivers challenge estimates of extreme river floods. Science Advances, 10(13), eadl4005.
- Wang et al. (2024). Distributed hydrological modeling with physics‐encoded deep learning: A general framework and its application in the Amazon. Water Resources Research, 60(4), e2023WR036170.
- Jiang et al. (2022). River flooding mechanisms and their changes in Europe revealed by explainable machine learning. Hydrology and Earth System Sciences, 26(24), 6339-6359.
News
Team
Associated group members