Home › Institute › FutureLabs › Artificial Intelligence in the Anthropocene

AI in Anthropocene

Artificial Intelligence in the Anthropocene - Machine Learning and data-driven Modelling in Earth System Science

Many phenomena in Earth system dynamics are complex in the sense that they emerge from nonlinear interactions of large numbers of processes across wide ranges of temporal and spatial scales. Modelling such phenomena on the basis of the underlying fundamental physical laws poses serious challenges, in particular in situations of strongly nonlinear behavior that may lead to abrupt state transitions, or if one is interested in the characteristics of extreme events.

In this Future Lab, hosted by PIK’s Research Department 4, we explore mathematical techniques to investigate and model complex Earth system processes with a strong focus on data-driven approaches.

Key research questions

How can techniques from Complexity Science and Machine Learning complement physical process-based approaches to

quantify the likelihood of abrupt transitions and extreme events in a warming Earth system?
improve predictions of extreme weather events on time scales of days to weeks?
assess the ecological impacts of a warming climate and changing extreme event characteristics?

Methods

We mainly employ methods from Complexity Science and Machine Learning such as

Complex Networks / Graphs for exploring dependencies in large datasets of climatic and ecosystem observables, to develop first hypotheses on underlying coupling mechanisms, and as a tool to coarse grain the data to extract the most relevant information
Bayesian inference for systematic calibration of physics-based low-order models that capture the key dynamics of the natural systems under study
Artificial Neural Networks to model (emergent) processes that are challenging to tackle with more traditional, primitive differential-equation-based approaches

Lead

Niklas Boers

Publications

Selected Publications

Contact

If you are interested in carrying out a BSc, MSc, or PhD project with us, please contact us to discuss possible topics.

Applications

We currently focus on the following areas of applications within the Earth system:

Abrupt climate transitions that have occurred in the Earth's long-term past, as evidenced in paleoclimate proxy records
Extreme events such as heat waves, droughts, and floods
Impacts of a warming climate and changing extreme-event characteristics on ecosystems, currently with a focus on boreal forests

Team

Niklas Boers (FutureLab leader, PIK and Technical University Munich - TUM)
Da Nian (Postdoc, PIK)
Maximilian Gelbrecht (Postdoc, Technical University Munich - TUM)
Takahito Mitsui (Postdoc, Technical University Munich - TUM)
Philipp Hess (PhD student, Technical University Munich - TUM)
Lana Blaschke (PhD student, PIK)
Keno Riechers (PhD student, PIK)
Maya Ben Yami (PhD student, Technical University Munich - TUM)
Alistair White (Phd student, Technical University Munich - TUM)
Jonas Pilot (MSc student, PIK)
Andreas Morr (MSc student, HU Berlin)
Dario Lepke (MSc student, LMU Munich)
Gabriele Pilz (scientific coordinator, PIK)

Former team members

Catrin Ciemer (Postdoc)
Jens Fohlmeister (Postdoc)
Nils Bochow (MSc student)
Christopher Böttner (MSc student)
Lucien Fumagalli (BSc student)
Tula Böschen (BSc student)
Eirik Myvol-Nisen (Postdoc)

Associated projects

EU Horizon 2020 project 'Tipping Points in the Earth System' (TiPES)
Freigeist Research group 'Predicting abrupt transitions and extremes in the Earth system' (FU Berlin, funded by the Volkswagen Foundation)
BMBF project ClimXtreme - Subproject B3.2: Spatial synchronization patterns of heavy precipitation events in Europe (SynXtreme)
EU Horizon 2020 International Training Network (ITN) 'CriticalEarth' (FU Berlin)

News & Media

Seasonal Prediction of Indian Summer Monsoon Onset with Neural Networks:

We have recently introduced a neural-network-based method to predict the Indian Summer Monsoon (ISM) onset at seasonal time scales (see here). From 1981 to 2021, the method hindcasts the ISM onset dates with an accuracy of 4.9 days (RMSE). For 2022, by March 11th our method predicts an ISM onset on May 26th (median), with an interquartile range from May 24th to 28th. For details, see here.