Machine Learning in Geosciences

  Superpixels in thin section Copyright: CGRE

We are involved in several projects developing and using machine learning methods to address geoscientific problems:

  • In the DFG TR32 project, we developed a combined Markov Random Field - Gaussian Mixture Model approach to link information from remote sensing data with geophysical measurements to determine soil patterns. Results have been published recently (JGR Biogeosciences, see "Selected Publications"). For more information, see Projects -> DFG TR32 SFB.
  • In the RWTH Seed-fund project "MINERALS", we address a simple, but highly important, issue for a wider use of ML in the field of geosciences: the development of open data bases with suitable training data sets. In this project, we combine expertise from thin section scanning and analysis (in collaboration with MaP and the institute of Structural Geology and the institute of Geology) with machine learning (in collaboration with the institute of Computer Vision). For more information, see Projects -> MINERALS.

In addition to these projects, we currently have several MSc and PhD projects at the link between machine learning and geosciences.