Previously Sabrina Berger* Sabrina Marcelo
Sabrina Marcelo Previously Sabrina Berger*
A lighthouse and a lone figure on a road beneath the arc of the Milky Way at night

Sabrina Marcelo

McWilliams Postdoctoral Fellow
Carnegie Mellon University

I study how the first supermassive black holes and their host galaxies formed and grew, and how the intergalactic gas around them was reionized. I combine cosmological simulations with statistical inference to interpret what JWST and upcoming 21-cm experiments can, and cannot, tell us about the early Universe.

Updates

Research

My research asks how the first supermassive black holes and their host galaxies formed, and how the intergalactic gas around them was reionized. Observations of this first billion years of our Universe are faint and shaped by instrumental effects. My approach is to forward model what telescopes actually record using cosmological simulations, and then use statistical inference to work back to physical quantities with quantified uncertainties.

  1. Early Black Holes and Their Host Galaxies

    Question. JWST can now detect the galaxies hosting quasars less than a billion years after the Big Bang, but the bright quasar can outshine its host. How well can host properties such as stellar mass be measured, and what do they imply about how early black holes and galaxies grew together?

    Approach. I generate mock JWST NIRCam images of simulated quasar hosts from the BlueTides cosmological hydrodynamic simulation and apply the same quasar (point-source) subtraction used on real observations. Building on these mocks, I developed a statistical method that uses thousands of simulated hosts to construct posteriors on host magnitudes corrected for subtraction biases (assuming a perfectly modelled PSF), and then fits the hosts' spectral energy distributions.

    Results. The first two quasar hosts detected with JWST, J2255+0251 and J2236+0032, lie on the BlueTides stellar-to-black-hole mass relation, and quasar subtraction works well at their luminosities (Berger et al. 2024, MNRAS). For JWST-observed quasar hosts more broadly, subtraction misestimates host magnitudes by at most 0.2 mag and stellar masses by less than 0.3 dex, so accounting for it does not resolve the apparent mismatch between black hole and host-galaxy growth at high redshift (Berger et al. 2026, MNRAS).

    Current. I am a Co-I on JWST Theory Program 10799 (PI: M. Marshall), interpreting JWST observations of high-z quasar environments with BlueTides. With MSc student Claire McKee, I am studying Little Red Dots in Rubin/LSST survey data.

    Papers: Berger et al. 2024, MNRAS (arXiv) · Berger et al. 2026, MNRAS (arXiv)

    Schematic titled 'How do supermassive black holes form?': massive black hole seeds (direct collapse, galaxy mergers, runaway star collisions) and light seeds from the first stars grow through Little Red Dots, quasars and other AGN into present-day host galaxies, along a timeline from redshift 20 to today.
    Proposed pathways for forming the first supermassive black holes: heavy and light seeds grow through phases such as Little Red Dots, quasars and other active galactic nuclei into the black holes in present-day galaxies. View full size

  2. In preparation

    Reionization and 21-cm Cosmology

    Question. During cosmic dawn and reionization, neutral hydrogen produces a faint 21-cm signal, seen in emission or absorption against the cosmic microwave background, that many experiments are now trying to detect through line-intensity mapping. How much can a 21-cm map alone tell us about the underlying distribution of matter, and where does that information run out?

    Approach. I use field-level Bayesian inference: Hamiltonian Monte Carlo, implemented in JAX on GPUs, samples the density field in Fourier space to give a per-pixel posterior on the matter density from a 21-cm brightness-temperature map.

    Results (on simulations). On simulated maps, the reconstruction reaches about 1 Mpc resolution. It is calibrated and nearly lossless in neutral regions, and degrades in a controlled way in ionized and high-density regions, where the 21-cm signal loses sensitivity to the density. Those regions call for complementary probes, such as galaxy maps from Roman or JWST. Manuscript in preparation.

    Context. This work grew out of my MSc thesis at McGill. I am Co-PI, with Adélie Gorce, of a 2025 French-Australian Science and Innovation Collaboration grant on the first galaxies with SKA-Low and AI, and a co-author of the SKA-Low 21-cm science chapter in Advancing Astrophysics with the SKA II.

    Links: MSc thesis · Acharya et al. 2026, AASKA II (arXiv)

    Four maps of a simulated field 128 Mpc across. Top row: the true matter density and its maximum a posteriori reconstruction, which shares the true field's large-scale structure. Bottom row: the corresponding 21-cm brightness temperature maps, with bright patches of neutral gas surrounded by dark ionized regions.
    Demonstration on a simulated 128 Mpc field. Top: the true matter density (1 + δ) and the maximum a posteriori (MAP) reconstruction from the 21-cm map alone. Bottom: the corresponding 21-cm brightness temperature; dark-blue regions are ionized and carry no 21-cm signal. View full size

  3. Radio Beam Calibration with GPS Satellites

    Question. Radio telescopes such as CHORD need accurate maps of their beams. Can the strong, continuous signals from GNSS satellites (e.g., GPS) be used to measure them?

    Approach. With Jonathan Sievers at McGill, I used a commercial GNSS receiver with the Deep Dish Development Array (D3A) at the Dominion Radio Astrophysical Observatory to map the 2D beam as satellites passed through its main lobe and sidelobes. Arianna Lasinski, an undergraduate I mentored, is second author.

    Results. Repeated measurements over three days probed several sidelobes, with the smallest day-to-day difference in the main lobe of 0.56 dB-Hz, and roughly matched simulations in shape. GNSS's continuous transmission, broad frequency coverage and frequent satellite passes make it a promising calibration method for CHORD, SKA-mid and the DSA-2000.

    Links: Berger et al. 2026, PASA (arXiv) · Code (GitHub) · MSc thesis

    A researcher holding a small receiver on a sunny rooftop next to a satellite dish.
    Early test data for this project were collected on the rooftops of McGill University and the UT Austin Physics Department. View full size

  4. Statistics of Spatial Variation

    Question. From the turbulent interstellar medium to the cosmic web, astronomers in different subfields describe spatial correlation with different statistics, which are often mathematically identical. How do they relate?

    Contribution. With Benjamin Metha (shared first authorship), I wrote a guide that defines the power spectrum, two-point correlation function, covariance function, semivariogram and structure functions, explains when each is useful, and shows how to translate results between them, so that methods can be shared across subfields.

    Paper: Metha & Berger 2025, PASP (arXiv)

Selected Publications

Selected first- and shared-first-author papers, published as S. Berger. Full publication list (ADS) · ORCID

  1. S. Berger, M. A. Marshall, J. S. B. Wyithe, T. Di Matteo, Y. Ni, S. M. Wilkins, M. Yue

    Quantifying biases in stellar masses of JWST high-z quasar host galaxies caused by quasar subtraction

    2026, MNRAS, 549, stag911

    A statistical method, built on thousands of simulated quasar hosts, shows that quasar subtraction biases JWST host stellar masses by less than 0.3 dex, so it does not explain the apparent overmassive black holes at high redshift.

    JournalarXiv

  2. S. Berger, A. Lasinski†, V. MacKay, E. Egan, D. Wulf, A. Chokshi, J. Sievers

    First Use of GPS Satellites for Beam Calibration of Radio Dish Telescopes

    2026, PASA, 43, e016

    Maps the beam of a radio dish, including several sidelobes, using GNSS satellite signals recorded with a commercial receiver.

    JournalarXiv

  3. B. Metha*, S. Berger*

    A “Rosetta Stone” for Studies of Spatial Variation in Astrophysical Data: Power Spectra, Semivariograms, Structure Functions, and More

    2025, PASP, 137, 073001

    A guide to translating between the spatial statistics used in different subfields of astronomy.

    JournalarXiv

* Shared first authorship. † Student I mentored.

Mentoring & Teaching

Research mentoring

Since 2020, I have mentored 15 students in research and professional development. Research students:

  • Juliana Ruiz Montoya, undergraduate research intern, RECA Internship Program (2026–present). Fast radio burst host galaxies in Rubin survey data. Student-led first-author manuscript in preparation.
  • Claire McKee, MSc student, University of Melbourne (2025–present). Little Red Dots in Rubin/LSST survey data.
  • Arianna Lasinski, undergraduate researcher, University of Toronto (2022–2026). GPS-based radio beam calibration and instrumentation. Second author on Berger et al. 2026, PASA; now an MSc student at the University of Barcelona.

Teaching

I have taught physics and computing at three universities and a software engineering bootcamp, emphasizing active learning and inclusive teaching.

  • University of Melbourne: tutor for Astrophysics (undergraduate, Semester 1 2026), Special Relativity (undergraduate, Semester 2 2023) and The Art of Scientific Computing (graduate, Semesters 1 and 2 2023).
  • McGill University: tutor for Astrophysics (honours undergraduate/graduate, Fall 2021), Signal Processing (Winter 2021 and 2022) and Computational Physics (honours undergraduate/graduate, Fall 2019); Linux/GitHub lecture at the McDonald Institute summer school (May 2021).
  • UC Berkeley: tutor for Python for Astronomy (Winter 2019); instructor for a Splash course on radio astronomy (February 2019).
  • Hackbright Academy: instructor in software engineering (2021–2022), delivering 25 hours of Python lectures each month and teaching more than 150 women and non-binary students to program.

Outreach

In this episode of the LIUniverse podcast (September 2025), I talk about my path into science and my research so far. My CV has a full list of my outreach.

astro[sound]bites

From 2022 to 2025, I co-hosted and managed astro[sound]bites, the podcast spin-off of astrobites, which turns current astrophysics research into accessible conversations for listeners at the undergraduate level. The podcast has more than 30,000 downloads; I hosted 36 episodes and co-led episodes on community-college pathways into astrophysics, Indigenous astronomy and mental health in astrophysics. Here are some of the episodes I'm most proud of:

Resources

Astrostatistics

Resources for Students and Researchers

About

Sabrina Marcelo smiling outdoors at dusk

I'm a second-generation Filipina-American from the San Francisco Bay Area, with roots in Puerto Princesa, Palawan. I went to UC Berkeley for a BA in Astrophysics (2018).

I then did an MSc in Physics at McGill University (2019–2022) with Adrian Liu and Jonathan Sievers, working on reionization and radio instrumentation, and a PhD in Physics at the University of Melbourne (2023–2026) with Stuart Wyithe. My dissertation was titled Beyond the Glare: Seeing High-z Supermassive Black Hole Host Galaxies with Simulations. Along the way, I was a data science intern at NASA Ames and a physicist intern at Meta, and co-hosted astro[sound]bites, a podcast about astronomy research.

I'm passionate about mentoring students from nontraditional backgrounds and expanding who gets to do research.

A note on my name: I have published as Sabrina Berger, and you may also know me as Sabrina Berger-Marcelo. I now go by Sabrina Marcelo.

CV

Academic timeline

  • McWilliams Postdoctoral Fellow, Carnegie Mellon University (2026–2029)
  • PhD in Physics, University of Melbourne (2023–2026)
  • Visiting PhD student, ANU and Swinburne
  • MSc in Physics, McGill University (2019–2022)
  • BA in Astrophysics, UC Berkeley (2018)

Industry

  • Physicist Intern, Meta (Facebook) (2021)
  • Data Science Intern, NASA Ames (2019)

Download CV (PDF)