SP1.3: Machine learning on radargrams

Subproject 1.3

Machine learning on radargrams

Information on ice thickness and internal structures of ice bodies (e.g. water table, isochrones, water pockets, and channels) from ground penetrating radargrams is to date often picked manually. Often only a specific target neglecting all other information is traced since existing contour-following algorithms in standard software like REFLEX do not provide consistent and reliable output. Within this doctoral project, we aim at using and modifying machine learning techniques from medical imaging as well as natural language processing (NLP) and apply those to glaciological radargrams. Ice thickness, bedrock topography, as well as internal structures, shall be mapped ideally at once after the respective pre-processing of the radargrams has occurred. Each radargram is composed of lines denoting different structures in the ice body. Algorithmically, this represents a segmentation problem in radargrams.
Datasets are available from airborne and ground-based low-frequency radar surveys of sites in the Alps and high Mountain Asia (BAdW), planned campaigns in Patagonia and alpine-type glaciers in Antarctica (FAU). Additional material for algorithm testing and training is available through an intense collaboration with AWI from polar surveys. Ideally, the developed algorithms will also be tested on the first survey data of the developed new multi-frequent radar in sub-project 1.1. Further links exist also to SP 2.3, and 3.1.

Co-PIs: A. Maier (FAU Informatics), T. Seehaus (FAU Geography), F. Navarro (U. Madrid)

Get to know our project affiliated PhD students

Marcel Dreier

Marcel Dreier


Hi, I am Marcel, and I studied Computer Science at Friedrich-Alexander-Universität Erlangen Nürnberg. I finished my master’s degree in August 2023, and I am currently a PhD student at the Pattern Recognition Lab at Friedrich-Alexander-Universität Erlangen Nürnberg under the supervision of Prof. Andreas Maier.
My research in the M3OCCA project focuses on the segmentation of glacier radar images using deep learning. Since glacier radar images are usually segmented by hand, geographers spend many hours dividing the glacier into different sections. Especially with the increase in data in recent years, this task has become very expensive. Hence, my research aims to alleviate that problem by using deep learning to speed up and automate the segmentation of glacier radargrams.
In my free time, I enjoy going for walks and bouldering.