Predicting the surface age of chondritic S-type asteroids using the space weathering features in reflectance spectra: Small data machine learning

¹ Department of Geosciences and Geography, University of Helsinki, Finland
² Department of Information and Computer Science, University of Helsinki, Finland
³ Astronomical Institute of the Czech Academy of Sciences, Czech Republic
⁴ Institute of Geology of the Czech Academy of Sciences, Czech Republic
⁵ School of Electrical Engineering, Aalto University, Finland

^★ Corresponding author

Received: 18 February 2025
Accepted: 26 May 2025

Abstract

Context. The surfaces of airless planetary bodies, such as S-type asteroids, undergo space weathering (SW) due to exposure to the interplanetary environment, resulting in alterations to their reflectance spectral features (e.g., spectral slope, albedo, and absorption band characteristics).

Aims. This study aims to estimate the surface age of S-, Sq-, and Q-type asteroids as a function of SW agents and dose by employing machine learning models.

Methods. Two models were developed: an ensemble model (combining a CNN, gradient-boosting regressor, K-nearest neighbor, extratree regressor, and random forest regressor) and a Gaussian process (GP) model. Both models were trained on published reflectance spectra of olivine, pyroxene, their mixtures, and chondritic meteorites, using SW conditions as independent variables and surface age at 1 AU as the dependent variable. Given the limited dataset, k-fold cross-validation was employed for model training. The models were further validated by applying them to S-, Sq-, and Q-type asteroids, evaluating their ability to capture two key trends: the SW progression across chondritic S-type asteroids and the relationship between asteroid size and surface age.

Results. Both models successfully identify relatively fresh surfaces in Q-type asteroids and mature surfaces in S-type asteroids, as well as younger surface ages for asteroids with diameters less than 5 km. However, the GP model exhibits higher variability in predictions for the asteroid dataset. While both models effectively capture relative surface age trends, limitations in data availability between 10³ and 10⁷ years hinder precise predictions of asteroid surface ages.

Conclusions. These models have significant potential for future applications, such as determining the surface age for individual asteroids and identifying asteroid families, offering valuable tools for advancing our understanding of asteroid evolution and SW processes.