Spectral fingerprints of pure and mixed minerals: Laboratory characterization and ML Integration

Kurzfassung

1. Introduction: Studies based on MESSENGER mission data have shown that the elemental composition of Mercury differs significantly from that of other rocky planets in the Solar System [1]. MESSENGER have revealed that Mercury’s surface has a low abundance of iron (Fe), which is typically found in higher concentrations on other rocky bodies [1, 2]. However, Mercury shows a higher concentration of magnesium, often mixed with different elements depending on the surface type [1, 2, 3]. Over the years, different studies have been conducted to understand the surface mineralogy of Mercury based on these initial understandings from the MESSENGER data. While we do have a good idea about the elemental abundances observed on Mercury, the exact mineralogical composition and its distribution remain poorly constrained. Although planned before MESSENGER entered orbit, the joint ESA-JAXA mission BepiColombo, launched in 2018, is equipped to address many of the new scientific questions raised by MESSENGER's findings. One of the instruments onboard BepiColombo is MERTIS (MErcury Radiometer and Thermal infrared Spectrometer) as part of the Mercury Planetary Orbiter (MPO) payload. MERTIS aims to investigate the mineral composition of Mercury and understand the planet’s thermal behavior [6]. It consists of an infrared spectrometer (TIS) with a spectral range of 7-14 μm with a resolution of 90 nm, and a radiometer (TIR) with a radiometric range of 7-40 μm, split into two bands - 8-14 μm and 7-40 μm [6]. In order to study the surface using MERTIS and bridge the gap towards understanding the mineral distribution, we are developing a spectral identification framework at the Planetary Spectroscopy Laboratories (PSL), DLR, Berlin, based on laboratory measurements of various Mercury analogs such as FeO-free enstatite, forsterite, albite etc. These measurements will serve as the foundation for a machine learning (ML) based identification algorithm, which will classify individual and mixed minerals [7] using distinctive spectral fingerprints identified from these spectra. 2. Dataset and Methodology: The dataset for the spectral identification framework will consist of the emissivity spectra measured at PSL using Mercury analogs such as magnesium-rich and FeO-poor minerals like enstatite, forsterite, olivine, labradorite, microcline, anorthoclase etc. [7]. In addition, several mixes of pure minerals with varying grain sizes (<25 µm, 25-63 µm and >125 µm) are being prepared to understand the influence of mixture and grain size on emissivity measurements at Mercury day-side temperature. PSL is equipped to measure emissivity spectra in vacuum (0.7 mbar) in the spectral range of MERTIS with temperatures from 100° to above 400° for a large suite of Mercury surface analogs. Out of the three spectrometers, one is equipped with an external chamber to measure the emissivity of solid samples (powder or slab). A shutter allows separating the spectrometer from the external chamber, that can be evacuated to the same pressure as the spectrometer [7]. To expand our library of emissivity spectra, we also aim to create synthetic spectra using various ratios of above-mentioned minerals using linear and non-linear mixing algorithms. These synthetic spectra will be cross-referenced with the laboratory measured spectra to calculate accuracy. The main goal for these different steps is to create a library of emissivity spectra dedicated to the MERTIS range and to automate and ease the process of identifying the mineral distribution on the surface of Mercury. 3. Preliminary results and future work: Emissivity measurements are currently being conducted at the Planetary Spectroscopy Laboratories, DLR, Berlin [7] on a broad range of Mercury analog minerals. In parallel, we are also developing an algorithm to extract and classify distinct spectral features from the measured spectra. We aim to use an unsupervised machine learning approach—specifically, using autoencoders—to detect key spectral features. This method will facilitate the identification of unique spectral features for minerals with different grain sizes and mixtures using both laboratory and synthetically generated emissivity spectra. To evaluate the algorithm’s performance, we will test it on unknown mixtures, prepared and measured in the lab, and assess its ability to correctly identify their mineralogical components.