Machine Learning Used for Meteorite Classification to Unlock Asteroid Composition Mysteries

Classification of asteroid spectra by analyzing meteorite spectra © lauritta - stock.adobe.com

Asteroids hold vital clues about the formation and history of our Solar System, but our understanding of their composition has been limited due to the rarity of asteroid sample-return missions. While remote sensing and asteroid imagery offer some insights, detailed interpretation of asteroid surface compositions remains challenging. A new study led by researchers from the Planetary Science Institute, Mount Holyoke College, and the University of Massachusetts Amherst, proposes an innovative solution: using machine learning (ML) to analyze asteroid spectra based on existing meteorite data (1).

Innovative Approach Using Meteorite Spectra
The study, authored by M. Darby Dyar, Sydney M. Wallace, Thomas H. Burbine, and Daniel R. Sheldon, takes a fundamentally different approach from previous asteroid taxonomies. Instead of relying solely on asteroid spectral data, which offers limited mineralogical information, the researchers turned to meteorite spectra, which provide direct evidence of their parent bodies. This meteorite-based classification allows for a robust taxonomy rooted in mineralogical, petrological, and chemical properties, unlike traditional asteroid classifications that often depend on more subjective spectral features (1).

The team utilized a dataset of 1,422 meteorite spectra from the Reflectance Experiment Laboratory (RELAB) at Brown University. They applied various ML algorithms, including logistic regression (LR), support vector machines (SVM), quadratic discriminant analysis (QDA), and kernel Fisher discriminant analysis (KFDA). The LR model proved the most effective, achieving a high classification accuracy of over 90%. This model was then applied to a set of 605 asteroid spectra, allowing the team to evaluate the distribution of compositions across the asteroid belt (1).

Spectroscopy and Key Wavelength Regions
Spectroscopy, the cornerstone of this study, was used to analyze the reflectance spectra of meteorites and asteroids across the visible to near-infrared (vis-NIR or VNIR) range (0.35–2.5 μm). The spectral features in this region are primarily influenced by mineral composition, particularly the presence of pyroxenes and olivine, which show characteristic absorption bands around 1.0 and 2.0 μm (1).

Normalization of the spectral data was crucial to ensure consistency across meteorite samples and asteroid predictions. The researchers explored multiple normalization techniques but ultimately found that normalizing spectra at 0.70 μm yielded the best results. This wavelength, slightly shifted from the historically used 0.55 μm solar maximum, enhanced the LR model’s ability to differentiate between meteorite classes and asteroid compositions (1).

Discover More: Spectroscopy of asteroids

Machine Learning Accuracy and Findings
Logistic regression (LR) was identified as the most accurate ML model, with a classification accuracy of approximately 92%. The LR model successfully grouped meteorites into eight distinct classes, allowing for clear correlations between specific meteorite types and asteroid parent bodies. For instance, the model confirmed known links between howardite-eucrite-diogenite (HED) meteorites and V-type asteroids, which are believed to originate from the asteroid designated as (4) Vesta. Similarly, the model matched enstatite chondrites with Xc-type asteroids and CV chondrites with K-type asteroids (1).

The study’s ML approach offers significant advantages over traditional classification methods. Most notably, it eliminates the subjectivity involved in visual inspection of spectra, which had previously hindered the accuracy of asteroid taxonomies like the Bus-DeMeo (BDM) classification system. Furthermore, the meteorite-based classification system allows for direct mineralogical insights into asteroid compositions, which is critical for understanding the building blocks of the Solar System (1,2).

Implications and Future Research
This new ML methodology paves the way for more accurate predictions of asteroid composition based on spectral data. By grounding asteroid classification in the well-established taxonomy of meteorites, the model offers a deeper understanding of the distribution of materials in the asteroid belt and the broader Solar System (1,2).

However, the researchers acknowledge the limitations of their approach, primarily the scarcity of spectral data for certain meteorite types, such as acapulcoites and CB chondrites. As more meteorites are discovered and spectral libraries expand, the classification model can be refined to improve accuracy further (1).

This study lays a solid foundation for future work, combining ML with spectroscopy to explore the mineralogical diversity of asteroids. By continuously refining the classification algorithms, scientists will be able to unravel the mysteries of asteroid composition, offering new insights into planetary formation and evolution (1).

This research marks a significant leap forward in asteroid classification and understanding the distribution of solid matter within the asteroid belt. By leveraging ML techniques and meteorite spectra, this study provides a robust framework for predicting asteroid compositions and furthering our knowledge of the early Solar System (1).

References

(1) Dyar, M. D.; Wallace, S. M.; Burbine, T. H.; Sheldon, D. R. A machine learning classification of meteorite spectra applied to understanding asteroids. Icarus 2023, 406, 115718. DOI: 10.1016/j.icarus.2023.115718

(2) Carruba, V.; Aljbaae, S.; Domingos, R. C.; Caritá, G.; Alves, A.; Delfino, E. M. D. S. Digitally filtered resonant arguments for deep learning classification of asteroids in secular resonances. MNRAS 2024, 531 (4), 4432–4443. DOI: 10.1093/mnras/stae1446