The project aims to develop novel deep learning architectures capable of learning the dynamics over space and time on Earth datasets. We base our effort on adapting stochastic video prediction algorithms to the properties of Earth datasets and condition the prediction on variables that are easy to predict by physical simulation models. We showcase an application of our method by predicting the landscapes of Earth for given environmental conditions as well as how these evolve over time.
Start/End:
2017/---
Project Leader:
Christian Requena-Mesa (FSU Jena, MPI Biogeochemistry)
Collaborators:
Joachim Denzler (FSU Jena)
Markus Reichstein (MPI Biogeochemistry)