PhD project offered by the IMPRS-gBGC in July 2021


Heterogeneous Earth observations and deep learning for modeling and understanding of land-atmosphere fluxes

Sophia Walther , Basil Kraft , Martin Jung , Markus Reichstein

Project description

The MPI-BGC based FLUXCOM initiative develops data-driven methods for simulating fluxes of carbon, water, and energy between the atmosphere and the land-surface and has successfully integrated satellite and site-level observations to create ensembles of global flux estimates (Jung et al., 2020, 2019, Bodesheim et al. 2018, Tramontana et al., 2016, Jung et al., 2010). The global simulations are used widely in Earth System sciences to address questions such as the impact of extreme events or climate change on terrestrial ecosystems, or disentangling anthropogenic and biogenic carbon contributions to the atmospheric concentrations.
Nowadays, a large range of space-borne instruments provide complementary information on the state of the land surface in different spectral regions and acquisition modes, as well as spatial, temporal, and spectral resolutions. The observational capabilities from space have strongly increased and diversified – offering unprecedented opportunities for a comprehensive diagnosis of terrestrial ecosystems with a variety of indicators. Jointly, this variety of satellite measurements can boost modelling performance and the spatiotemporal detail of the simulations. This is of particular relevance in ecosystems that are very heterogeneous and / or undergo strong management. Examples are global crop productivity, selective wood harvest and forest disturbances, or high latitude ecosystems, all of which are notoriously challenging to represent realistically in FLUXCOM but are of high scientific interest as well as of relevance for society, food security, and policy.
At the same time, developments in machine learning open new avenues for the integration of diverse data sources. Novel deep learning architectures that exploit spatio-temporal interactions have the potential to bring flux predictions to the next level.
In this PhD project we seek to explore new ways to synergistically exploit diverse sets of Earth Observation data in FLUXCOM. The candidate will develop methods to explicitly or implicitly account for heterogeneities in acquisition properties of space-borne instruments (e.g., resolution, observation geometries, and overpass times) using machine learning, and to use the resulting flux estimates on targeted regions and specific scientific questions. Depending on the candidate's skills and interest, the project can be focused on ecological or methodological aspects. The thematic orientation is to be defined; possible research questions are:

Ecological questions:
  • How can novel data sources help to better account for and understand the impact of natural disturbances on ecosystem fluxes?
  • How can we account for human management and interventions in flux modeling using Earth observation data and improved simulations, e.g., for crop productivity?
  • Which ecosystem processes can be resolved across scales and sensor resolutions and what are the ‘blind spots’ that we miss, especially in fragmented landscapes?

Methodological questions:
  • Where and when do Earth observation data streams individually and synergistically matter most for flux predictions?
  • Data fusion: combine datasets with different spatial and temporal resolution for time-series modeling (e.g., attention-based models)
  • Super-resolution imaging: ML-based down-scaling of coarse-resolution satellite products
  • Transfer learning and self-supervision: knowledge transfer from related tasks / data
Working group

Global Diagnostic Modeling (GDM)


We are looking for a motivated candidate with a strong interest in data analysis, machine learning, ecology, and climate. The successful candidate will work in close collaboration with an international and diverse research team. More specifically, the requirements are:
  • A Master's degree (or equivalent) in computer science, geoinformatics, remote sensing, or similar.
  • Computational skills: programming skills (e.g., Python, R, Matlab), processing and analyzing large data sets, machine learning, statistical analysis.
  • Experience with deep learning frameworks (e.g., PyTorch) is an asset, but not required.
  • Self-driven personality able to work both independently and in a team.
  • Excellent oral and written communication skills in English.
The Max Planck Society seeks to increase the number of women in those areas where they are underrepresented and therefore explicitly encourages women to apply. The Max Planck Society is committed to increasing the number of individuals with disabilities in its workforce and therefore encourages applications from such qualified individuals.


Jung et al. (2020), Biogeosciences
Jung et al. (2019), Scientific Data
Bodesheim et al. (2018), ESSD
Tramontana et al. (2016)
Jung et al. (2010), J. Geophys. Res.

>> more information about the IMPRS-gBGC + application