arXiv subjects

C. Cantero

Publications and source records attributed to C. Cantero.

6 recordsLinked to original sources

arXiv2026

Machine Learning as a Transformative Tool for (Exo-)Planetary Science

The exploration of planetary bodies in our Solar system and beyond relies on the processing and interpretation of large, spatio-temporally inconsistent, and heterogeneous datasets. Recent advances in machine learning (ML) provide unprecedented opportunities to address many fundamental challenges posed by these heterogeneous and hyper-dimensional datasets. This review chapter highlights innovative ML methodologies that were developed and used by NCCR PlanetS members to address three overarching challenges in (exo)planetary science. The first challenge is sequence modelling, which encompasses the intricate analysis of one-dimensional data such as time series of radial velocities and light curves, among other examples. Secondly, there is pattern recognition that involves studying correlations, leveraging convolutional neural networks for feature extraction, mapping and cross correlation among other examples., anomaly detection through variational autoencoders, and unsupervised clustering of mass spectrometric data. Lastly, there are generative models and emulation-based Bayesian analysis, which encompass the development of predictive models for planetary interior structure, employing Deep Neural Networks to understand planet formation mechanisms. These innovative ML methodologies herald a paradigm shift in the processing of data and numerical models that represent inherent challenges in planetary and exoplanetary science, paving the way for revolutionary discoveries and ideas in this field.

astro-ph.EP↗

arXiv2022

Exoplanet Imaging Data Challenge, phase II: Characterization of exoplanet signals in high-contrast images

Today, there exists a wide variety of algorithms dedicated to high-contrast imaging, especially for the detection and characterisation of exoplanet signals. These algorithms are tailored to address the very high contrast between the exoplanet signal(s), which can be more than two orders of magnitude fainter than the bright starlight residuals in coronagraphic images. The starlight residuals are inhomogeneously distributed and follow various timescales that depend on the observing conditions and on the target star brightness. Disentangling the exoplanet signals within the starlight residuals is therefore challenging, and new post-processing algorithms are striving to achieve more accurate astrophysical results. The Exoplanet Imaging Data Challenge is a community-wide effort to develop, compare and evaluate algorithms using a set of benchmark high-contrast imaging datasets. After a first phase ran in 2020 and focused on the detection capabilities of existing algorithms, the focus of this ongoing second phase is to compare the characterisation capabilities of state-of-the-art techniques. The characterisation of planetary companions is two-fold: the astrometry (estimated position with respect to the host star) and spectrophotometry (estimated contrast with respect to the host star, as a function of wavelength). The goal of this second phase is to offer a platform for the community to benchmark techniques in a fair, homogeneous and robust way, and to foster collaborations.

astro-ph.IM↗

arXiv2021

Exoplanet Imaging Data Challenge: benchmarking the various image processing methods for exoplanet detection

The Exoplanet Imaging Data Challenge is a community-wide effort meant to offer a platform for a fair and common comparison of image processing methods designed for exoplanet direct detection. For this purpose, it gathers on a dedicated repository (Zenodo), data from several high-contrast ground-based instruments worldwide in which we injected synthetic planetary signals. The data challenge is hosted on the CodaLab competition platform, where participants can upload their results. The specifications of the data challenge are published on our website. The first phase, launched on the 1st of September 2019 and closed on the 1st of October 2020, consisted in detecting point sources in two types of common data-set in the field of high-contrast imaging: data taken in pupil-tracking mode at one wavelength (subchallenge 1, also referred to as ADI) and multispectral data taken in pupil-tracking mode (subchallenge 2, also referred to as ADI mSDI). In this paper, we describe the approach, organisational lessons-learnt and current limitations of the data challenge, as well as preliminary results of the participants submissions for this first phase. In the future, we plan to provide permanent access to the standard library of data sets and metrics, in order to guide the validation and support the publications of innovative image processing algorithms dedicated to high-contrast imaging of planetary systems.

astro-ph.IM↗

arXiv2019

A census of $ρ$ Oph candidate members from Gaia DR2

The Ophiuchus cloud complex is one of the best laboratories to study the earlier stages of the stellar and protoplanetary disc evolution. The wealth of accurate astrometric measurements contained in the Gaia Data Release 2 can be used to update the census of Ophiuchus member candidates. We seek to find potential new members of Ophiuchus and identify those surrounded by a circumstellar disc. We constructed a control sample composed of 188 bona fide Ophiuchus members. Using this sample as a reference we applied three different density-based machine learning clustering algorithms (DBSCAN, OPTICS, and HDBSCAN) to a sample drawn from the Gaia catalogue centred on the Ophiuchus cloud. The clustering analysis was applied in the five astrometric dimensions defined by the three-dimensional Cartesian space and the proper motions in right ascension and declination. The three clustering algorithms systematically identify a similar set of candidate members in a main cluster with astrometric properties consistent with those of the control sample. The increased flexibility of the OPTICS and HDBSCAN algorithms enable these methods to identify a secondary cluster. We constructed a common sample containing 391 member candidates including 166 new objects, which have not yet been discussed in the literature. By combining the Gaia data with 2MASS and WISE photometry, we built the spectral energy distributions from 0.5 to $22\microm$ for a subset of 48 objects and found a total of 41 discs, including 11 Class II and 1 Class III new discs. Density-based clustering algorithms are a promising tool to identify candidate members of star forming regions in large astrometric databases. If confirmed, the candidate members discussed in this work would represent an increment of roughly 40% of the current census of Ophiuchus.

astro-ph.EP↗

arXiv2019

Random Forest identification of the thin disk, thick disk and halo Gaia-DR2 white dwarf population

Gaia-DR2 has provided an unprecedented number of white dwarf candidates of our Galaxy. In particular, it is estimated that Gaia-DR2 has observed nearly 400,000 of these objects and close to 18,000 up to 100 pc from the Sun. This large quantity of data requires a thorough analysis in order to uncover their main Galactic population properties, in particular the thin and thick disk and halo components. Taking advantage of recent developments in artificial intelligence techniques, we make use of a detailed Random Forest algorithm to analyse an 8-dimensional space (equatorial coordinates, parallax, proper motion components and photometric magnitudes) of accurate data provided by Gaia-DR2 within 100 pc from the Sun. With the aid of a thorough and robust population synthesis code we simulated the different components of the Galactic white dwarf population to optimize the information extracted from the algorithm for disentangling the different population components. The algorithm is first tested in a known simulated sample achieving an accuracy of 85.3%. Our methodology is thoroughly compared to standard methods based on kinematic criteria demonstrating that our algorithm substantially improves previous approaches. Once trained, the algorithm is then applied to the Gaia-DR2 100 pc white dwarf sample, identifying 12,227 thin disk, 1,410 thick disk and 95 halo white dwarf candidates, which represent a proportion of 74:25:1, respectively. Hence, the numerical spatial densities are $(3.6\pm0.4)\times10^{-3}\,{\rm pc^{-3}}$, $(1.2\pm0.4)\times10^{-3}\,{\rm pc^{-3}}$ and $(4.8\pm0.4)\times10^{-5}\,{\rm pc^{-3}}$ for the thin disk, thick disk and halo components, respectively. The populations thus obtained represent the most complete and volume-limited samples to date of the different components of the Galactic white dwarf population.

astro-ph.SR↗

arXiv2018

A White Dwarf catalogue from Gaia-DR2 and the Virtual Observatory

We present a catalogue of 73,221 white dwarf candidates extracted from the astrometric and photometric data of the recently published Gaia DR2 catalogue. White dwarfs were selected from the Gaia Hertzsprung-Russell diagram with the aid of the most updated population synthesis simulator. Our analysis shows that Gaia has virtually identified all white dwarfs within 100 pc from the Sun. Hence, our sub-population of 8,555 white dwarfs within this distance limit and the colour range considered, $-\,0.52<(G_{\rm BP}-G_{\rm RP})<0.80$, is the largest and most complete volume-limited sample of such objects to date. From this sub-sample we identified 8,343 CO-core and 212 ONe-core white dwarf candidates and derived a white dwarf space density of $4.9\pm0.4\times10^{-3}\,{\rm pc^{-3}}$. A bifurcation in the Hertzsprung-Russell diagram for these sources, which our models do not predict, is clearly visible. We used the Virtual Observatory tool VOSA to derive effective temperatures and luminosities for our sources by fitting their spectral energy distributions, that we built from the UV to the NIR using publicly available photometry through the Virtual Observatory. From these parameters, we derived the white dwarf radii. Interpolating the radii and effective temperatures in hydrogen-rich white dwarf cooling sequences, we derived the surface gravities and masses. The Gaia 100 pc white dwarf population is clearly dominated by cool ($\sim$ 8,000 K) objects and reveals a significant population of massive ($M \sim 0.8 M_{\odot}$) white dwarfs, of which no more than $\sim$ $30-40 \%$ can be attributed to hydrogen-deficient atmospheres, and whose origin remains uncertain.

astro-ph.SR↗