PhD project offered by the IMPRS-gBGC in July 2026

Hybrid Machine Learning for Flood Prediction across Space and Time

Shijie Jiang , Alexander Brenning , Markus Reichstein

Project description

Floods are among the most damaging environmental hazards, yet predicting them where they matter most remains difficult. Most of the world's catchments are ungauged, and the records we do have are often short. Climate change and land surface alterations make the past an increasingly unreliable guide to the future. Purely data-driven models have made remarkable progress by learning across many catchments at once, but they remain constrained to the range of conditions they have seen, and the cases we care about most, unprecedented extremes and a shifting climate, lie outside that range by construction.

This project will advance hybrid, physics-aware machine learning for flood prediction that aims to generalize beyond observed conditions. Building on ADELM (https://adelm.org/), our differentiable ecohydrological land surface model developed openly in the group, the project will develop a model chain that runs from land-surface processes through the river network to flood inundation, trained end to end against observations. A key focus will be generalization across space and time, including ungauged regions, non-stationary climatic and land-surface conditions, and rare extreme events. The successful candidate will work at the interface of hydrology, land surface modeling, and machine learning. The project offers the opportunity to develop hybrid and differentiable modeling techniques, to combine process understanding with large observational datasets, and to improve the reliability with which we can anticipate floods under environmental changes.

Working group & collaboration

The successful candidate will work in the Biogeochemical Integration department at the Max Planck Institute for Biogeochemistry and will also be affiliated with Friedrich Schiller University, Jena. The working group offers long-standing expertise in ecohydrology, environmental systems modeling, and hybrid and interpretable machine learning. The research connects to ongoing work in the team on AI generalizability in non-stationary environmental regimes and hydro-climatic extremes within the GENAI-X project (https://www.genai-x.uni-jena.de/). The PhD candidate will engage closely with the ELLIS Unit Jena as part of the European Lab for Learning and Intelligent Systems (ELLIS), benefiting from a strong international machine learning research network. For further information, please contact Shijie Jiang.

Requirements

Applications to the IMPRS-gBGC are open to well-motivated and highly-qualified students from all countries. Prerequisites for this PhD project are:

Master's degree in a relevant field, on the environmental side (Earth system science, hydrology, remote sensing, climate, ecology) or the quantitative side (computer science, data science, physics, or similar), or ideally a combination of both.
Hands-on experience building or modifying models, going beyond running existing tools out of the box. This could be a machine learning model whose architecture or training you designed yourself, a process-based environmental model you modified or extended, or a hybrid of the two. Strong candidates without this experience but with the foundations and motivation to take it on are encouraged to make that case.
Good programming skills, preferably in Python.
Fluency in spoken and written English, and the ability to communicate scientific ideas clearly within an interdisciplinary team.

The following would be strong assets, though not required:

Familiarity with differentiable programming and automatic differentiation (e.g., PyTorch, JAX).
Experience with hydrologic, land-surface, or routing/hydrodynamic models.
Comfort with numerical methods for dynamical systems and with messy geospatial or time-series data.

The Max Planck Society (MPS) strives for gender equality and diversity. The MPS aims to increase the proportion of women in areas where they are underrepresented. Women are therefore explicitly encouraged to apply. We welcome applications from all fields. The MPS has set itself the goal of employing more severely disabled people. Applications from severely disabled persons are expressly encouraged.

Reference

Wang, C., Jiang, S., Zheng, Y., 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.