Scaling uncertainties on asteroid characteristics to prepare datasets for machine learning
Introduction
The exploration of asteroids is currently driven by three fields of interest: the knowledge of the Solar System Formation (SSF), Planetary Defence (PD) and Resources Prospecting (RP) to prepare its utilisation. Machine learning (ML) methodologies appear very promising in that context, considering the expected large datasets (soon measured in Exabytes in multiple astronomy experiments). While databases currently present large heterogeneities preventing the full implementation of ML, it is only a matter of time before algorithmic datasets play a dominant role in the area of asteroid sciences
This paper focuses on a preliminary normalisation of the quantification of asteroids characteristics to prepare the datasets for ML. Section 2 presents the motivations for asteroid characterisation for a high number of targets. Section 3 details the currents and foreseen uses of ML in asteroids science. Section 4 presents the state-of-the-art methodologies used to determine different characteristics of asteroids (size/shape, orbital dynamics, mass/density, spin, internal structure and composition) and introduces proposals for characterisation factors. Finally, Section 5 concludes on how the characterisation factors can be used to create datasets for ML, to verify the efficiency of its predictions and states on the value and limitations of the proposed methodology, as well as the next identified steps.
Section snippets
Knowledge of the formation and evolution of the Solar System
There are many unanswered questions about SSF, as identified in (Morbidelli and Raymond, 2016), such as “1) the structure and evolution of proto-planetary disks; 2) the growth of the first planetesimals; 3) orbital migration driven by interactions between proto-planets and gaseous disk; 4) the origin of the Solar System’s orbital architecture; and 5) the relationship between observed super-Earths and our own terrestrial planets”. Small bodies are remnants of the proto-planetary disk that the
ML in asteroid science
ML or Artificial Intelligence (AI) is mainly used for classification, regression, clustering, forecast generation and reconstruction, discovery and insight. (Fluke and Jacobs, 2019) describes the state of the art of ML used with modern astronomical datasets. ML is usually classified as supervised or unsupervised:
- –
Supervised ML, relying on a training dataset (typically 70% to 90% of the overall data), a validation dataset and a testing dataset. Supervised ML can be used to perform classification
Asteroids characterisation: What do we know, and how?
There is a gap between the rate of asteroid discoveries (around one hundred each months) and characterisation (Galache et al., 2015), that is done using remote observation campaigns, fly-bys, or through dedicated missions.
Asteroid characterisation is made using photometric, spectroscopic, ephemeris and more rarely, imaging datasets. The following sub sections are a review of the different characteristics of asteroids, the methods that are currently being used to determine each characteristic
Conclusions
Preliminary elements for characterisation scaling factors are defined in this article, in order to define the level of knowledge available about a specific characteristic of an asteroid. Moreover, application specific characterisation factors have been proposed in Section 4.9. While the maximum TCF of 36 is currently not reached for any asteroid, it is reasonable to think that it could be obtained for a limited number of bodies in the coming decades, notably in the frame of RP and mining. A
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledgements
The authors would like to thank B. Segret for his support, R. Tsai for proofreading and the two anonymous reviewers for their constructive and helpful comments.
References (38)
Density of asteroids
Planet. Space Sci.
(2012)The long-term scientific benefits of a space economy
Space Policy
(2016)How many ore-bearing asteroids?
Planet. Space Sci.
(2014)- et al.
The need for speed in Near-Earth Asteroid characterization
Planet. Space Sci.
(2015) On the revision of radiometric albedos and diameters of asteroids
Icarus
(1997)- et al.
Direct observations of asteroid interior and regolith structure: science measurement requirements
Adv. Space Res.
(2018) - et al.
Mission opportunities for the flight validation of the kinetic impactor concept for asteroid deflection
Acta Astronautica
(2014) - et al.
Hidden mass in the asteroid belt
Icarus
(2002) - et al.
Asteroids: recent advances and new perspectives
Asteroids IV
(2015) Deflection of near-Earth asteroids by kinetic energy impacts from retrograde orbits
Planet. Space Sci.
(2004)
Practical applications of asteroidal ISRU in support of human exploration
Iron alteration minerals in the visible and near-infrared spectra of low-albedo asteroids
Icarus
The asteroid lightcurve database
Icarus
Silhouette-Based 3D Shape Reconstruction of a Small Body from a Spacecraft
Shaping asteroid models using genetic evolution (SAGE)
MNRAS
The Yarkovsky and YORP effects: Implications for asteroid dynamics
Annu. Rev. Earth Planet. Sci.
The flux of small near-Earth objects colliding with the Earth
Nature
Machine-learning identification of asteroid groups
MNRAS
Machine learning classification of new asteroid families members
MNRAS
Cited by (2)
Artificial Intelligence for Trusted Autonomous Satellite Operations
2024, Progress in Aerospace SciencesFeedforward neural network based time-varying state-transition-matrix of Tschauner-Hempel equations
2022, Advances in Space Research