Speaker
Description
BepiColombo is the joint European Space Agency (ESA) and Japan Aerospace Exploration Agency (JAXA) mission to Mercury. It carries the Mercury Radiometer and Thermal Infrared Spectrometer (MERTIS), the first instrument designed to obtain thermal-infrared spectra of Mercury at wavelengths of 7–14 µm. During the fifth Mercury flyby in December 2024, MERTIS acquired more than 1.4 million surface emissivity spectra. Global orbital observations are expected after Mercury orbit insertion and the start of nominal science operations in 2027.
At AGH University of Krakow, we will train and evaluate machine-learning algorithms for the mineralogical and geological interpretation of MERTIS data. The algorithms will complement existing laboratory-to-orbit mineral identification approaches for MERTIS by focusing on spectral classification against Planetary Spectroscopy Laboratory (PSL) emissivity spectra. We will train models based on spectral matching, dimensionality reduction, and unsupervised clustering on thermally corrected emissivity spectra. This work does not involve training large language models on spectral data. Given the sensitivity of mission-related data, we consider Polish language models such as Bielik, which can be run fully locally, to support code production, documentation, and agents that call the spectral algorithms. Computing resources at ACK Cyfronet AGH (PLGrid), including the Helios supercomputer, will allow the algorithms to be trained and tested on flyby and orbital-scale datasets. The resulting algorithms and evaluation procedures will support the use of MERTIS observations to investigate the composition and geological history of Mercury’s surface.