LINCC Frameworks Incubator Awardees

The LINCC Frameworks Incubator Program supports teams of researchers to expand early-stage analysis software being developed as a part of LINCC Frameworks using their own scientific investigations. Scroll through this page to learn more about each of the Incubator Projects.

Interested in your own incubator? View the Call for Proposals for more information on how to apply.

Funded Incubator Projects:

Supernova Template Fitting for the Age of LSST

Primary Investigator: Kaylee de Soto, Harvard University

headshot of incubator awardee Kaylee De Soto

Transient astrophysics probes the evolution of the Cosmos on truly human timescales. Of particular interest are supernovae, the explosive death of stars. The Vera Rubin Observatory’s LSST will discover thousands of supernovae (and other transients) every night. Given these high event rates, rapid inference tools are essential in quick classification and determining appropriate follow-up strategies. In this project, we will enable real-time fitting of transient-like light curves discovered with the Rubin Observatory using a simple parametric model. We will compare the use of variational inference, MCMC, and nested sampling in terms of accuracy, precision, and computational cost/speed for these models. Using a neural network, the posteriors of the fitted parameters of our empirical model will be mapped to physical parameters. This method of feature extraction will be incorporated as a filter in the ANTARES Broker.

Partnering with the LINCC Frameworks team will enable us to scale up these inference techniques to LSST data rates and provide a flexible community tool. Via module design, we hope our framework can incorporate user-defined models and new inference techniques as they are developed. This work will require a combination of algorithmic, scalability, productionization, and machine learning experience.

Publications resulting from this project:

“Superphot+: Real-time Fitting and Classification of Supernova Light Curves.” Kaylee M. de Soto et al 2024 ApJ 974 169. [IOPscience , arXiv]

Optimizing an LSST Solar System Simulator

Primary Investigator: Meg Schwamb, Queen’s University Belfast

The Vera C. Rubin Observatory Legacy Survey of Space and Time (LSST) will discover more than six million new Solar System bodies. This is an order of magnitude more objects than are currently known today in each of the Solar System’s small body reservoirs. LSST will go beyond just discovery; with a 10-year baseline, the survey will be able to measure broad-band optical colors and phase curves and capture episodes of cometary activity, orbit changes, rotational breakup events, and rotational brightness variations. Planetesimals are the bricks and mortar left over after the construction of planets. Their compositions, shapes, densities, rotation rates, and orbits help reveal their formation history, the conditions in the planetesimal-forming disk, and the processes active in the Solar System today. LSST will transform our current view of the Solar System and let us peer back into the Solar System’s past like never before.

The LSST Solar System Science Collaboration (SSSC) has identified key software products/tools that the Rubin user community must develop to achieve the planetary community’s LSST science goals. Near the top of the SSSC’s software roadmap is a Solar System survey simulator to enable comparisons of model small body orbital and size/brightness distributions to LSST discoveries. For the past several years, we have been developing an open-source community LSST Solar System Survey Simulator that takes a model Solar System small body population and uses the pointing history, observation metadata, and expected Rubin Observatory detection efficiency to output what LSST should find so that the numbers and types of simulated detections can be directly compared to the number and types of real small bodies found in the actual LSST survey.  We have developed use cases/user stories and a design for the overall code architecture that works for all Solar System populations that the SSSC/planetary community will want to compare to models and orbital/size/compositional maps, but we are struggling with scaling up our algorithms to LSST data rates and optimizing the code. Partnering with experts in data structures, databases, scalability, productionization, and code architecture through this LINCC Frameworks Incubator will enable us to truly make the LSST Solar System Survey Simulator a real open-source, community-wide tool.

Publications resulting from this project:

“Controlling Randomization in Astronomy Simulations.” Megan E. Schwamb et al 2024 Res. Notes AAS 8 25. [IOPscience].

DeepDISC LSST: photo-z

Primary Investigator: Grant Merz, University of Illinois Urbana Champaign

With the help of the LINCC Frameworks team, we have implemented our image-based model, DeepDISC, within RAIL [Github] and improved the original code base [Github]. Our code follows a modular design and with community-oriented tools to promote the use of image-based photo-z methods.  We will test our photo-z model with DESC DC2 data and compare to other codes within RAIL.  Further validation on real and simulated data will be included in future work.

With the first light of the Vera C. Rubin Observatory on the horizon, astronomers will be faced with unprecedented amounts of data at never-before-seen depths for ground-based observations. Photometric redshift, or photo-z, estimation is an important task in survey pipelines for cosmological analyses, as spectroscopic redshift measurements are too costly for large surveys such as LSST. Existing frameworks for photo-z estimation include template fitting methods, as well as machine/deep learning models. Image-based models are able to incorporate colors as well as morphology information into predictions.  Our method, DeepDISC, uses instance segmentation models to simultaneously detect, deblend, and estimate source redshifts using images as input. It is an efficient method for extracting object features in large cutouts, and can be applied to a variety of image sizes with varying source density.

Publications resulting from this project:

“DeepDISC-photoz: Deep Learning-Based Photometric Redshift Estimation for Rubin LSST.” Preprint submitted Nov 2024. [arXiv]

Integrating Robust Cross-Matching from the LSST: UK into the LINCC Frameworks

Primary Investigator: Tom J Wilson, University of Exeter

The Vera C. Rubin Observatory’s Legacy Survey of Space and Time (LSST) will revolutionize many aspects of astronomy all by itself, but there are important outstanding questions for which the real benefits will come from the combining of LSST data with other datasets. However, the LSST will probe such depths, and unprecedented levels of source crowding, in the night sky that standard cross-matching algorithms, used to determine counterparts between two photometric catalogs, will begin to break down. The sheer density of detected LSST sources across the entire southern sky means that there will be something like, at minimum, two and upwards of ten randomly placed objects nearby each and every LSST source, which will confuse counterpart identification. Worse still, the levels of object crowding observed, especially in the Galactic plane, mean that the positions of objects recorded in catalogs begin to be affected by even fainter, undetected sources hiding in their footprints. To that end, as part of an in-kind contribution, we have developed novel cross-matching algorithms tuned for handling the crowded sky that the Rubin Observatory will reveal.

This Incubator program’s goal was the integration of cross-matches from our codebase, macauff [Github], into the LINCC Frameworks software, LSDB [Github], offering robust counterparts in the crowded LSST sky. Working with LINCC Frameworks experts in catalog storage, database partitioning and querying, and scalable spatial analysis, we created an efficient and effective way to run, store, and serve this new method of determining counterpart assignments, lsdb-macauff [Github], to the wider Rubin community, maximizing the scientific return of LSST.

Developing the LePhare photometric redshift code to improve validation, robustness, usability and performance at scale

Primary Investigator: Raphael Shirley, Max Planck Institute for Extraterrestrial Physics

headshot of astronomer Raphael Shirley

In order to study galaxies we need to know how far away they are. In practice this means measuring their redshift either using spectroscopy or photometric measurements of colour to estimate them. These photometric redshifts can be measured using various methods including fitting template spectra to the colours. This will be crucial to doing science with the Vera C. Rubin Observatory data. In this incubator we worked on the LePHARE code to improve its robustness and applicability to Rubin data. Le phare is French for lighthouse and comes from the acronym PHotometric Analysis for Redshift Estimation. I think a lighthouse is a great metaphor because it evokes navigating the night sky like mariners using the stars. In this incubator we made the code much easier to install and use and did a lot of work implementing the latest standards in software development. We also integrated the LePHARE code to the Redshift Assessment and Infrastructure Layers that will be used to calculate redshifts at scale for all the Rubin detected galaxies.

The LINCC Frameworks engineers led by Drew Oldag brought their expertise as professional software developers to the project and provided fresh insight into the design process. This meant that we could harness the talents from the scientists and engineers together and in just three months really improved various aspects of the code. I think this will help us to promote the code by helping new users to get started. I also think that it will be great for the Legacy Survey of Space and Time in particular by making sure we have robust estimates of redshift for every object in the huge catalogues it will produce. This will be crucial for all the extragalactic astronomy that will be done with this incredible telescope.

Survey Masks and Ultrafast Correlation Functions for the Astronomical Community

Primary Investigator: Emilio Donoso, Instituto de Ciencias Astronómicas, de la Tierra y del Espacio (ICATE-CONICET) & Universidad Nacional de San Juan

LSST offers an exciting opportunity to characterize how galaxies cluster in the universe. A huge volume sampled at unprecedented depth and carefully calibrated photometric redshifts, will allow the computation of accurate estimates of the spatial distribution of sources by means of the two-point auto-correlation and cross-correlation functions (CFs) for samples with billions of objects. Correlation functions are powerful statistical tools across many domains in astronomy ranging from cosmology, galaxy evolution, unification models of active galactic nuclei, galactic structure, and even the multiplicity of star-forming regions, to name a few.

In this project, we will develop the software framework to enable such computation at LSST-scale. The first outstanding issue we will address concerns angular masks. CFs require the definition of precise masks highlighting the effective area accessible by the survey at any given time, accounting for regions around bright stars, bad photometry, high galactic extinction, and zones without observations. Masks are relevant not only to CFs, but also to other important quantities such as the luminosity function and even simple cross-matching between surveys. We will develop procedures to generate these masks efficiently considering the constraints astronomers will face when accessing the sheer volume of LSST data. The second step involves the actual pair counting of sources as a function of scale. Our team will join forces with a group of highly skilled LINCC Frameworks scientists to integrate the state-of-the-art, fast counting algorithms we have developed, with the LINCC Frameworks database libraries that will store, partition and serve the vast LSST catalogs.

photo-D: Estimating Stellar Distances with the LSST’s Broad-Band Photometry

Primary Investigator: Lovro Palaversa, Ruđer Bošković Institute

headshot of a man smiling with glasses

The Large Synoptic Survey Telescope (LSST) will, for the first time in history, catalog more Milky Way stars than there are living people on Earth—on the order of 10-20 billion, depending on model assumptions. To map the Milky Way in three dimensions, distances to these stars must be accurately estimated. The Photo-D distance estimation method produces color-based estimates of distances for billions of stars with LSST measurements. Additionally, this fully Bayesian procedure will also produce estimates of stellar parameters such as metallicity and surface gravity, and interstellar dust extinction along the line of sight to each star. These additional LSST data products will enable studies of the Milky Way ranging from tests of models for its formation and evolution to the search for stellar streams, which are excellent probes of dark matter distribution.

LINCC Frameworks Incubator grant enabled us to ensure the computational tractability of our method. We successfully decreased computation time per star by a factor of 5–10x and developed GPU-compatible code, significantly enhancing performance. Beyond optimizing the codebase, substantial gains were achieved by integrating HATS [ReadTheDocs] and LSDB [LSDB.io] as core components of the pipeline. These improvements not only enhance computational efficiency but also simplify future code maintenance and facilitate the addition of new modules.

Publications resulting from this project:

“PhotoD with LSST: Stellar Photometric Distances Out to the Edge of the Galaxy.” Lovro Palaversa et al 2025 AJ 169 119. [IOPscience , arXiv]

Building an Anomaly Detection Recommendation System for Novel Transients in Early LSST Data

Primary Investigator: Amanda Wasserman, University of Illinois Urbana Champaign

The Vera Rubin Observatory’s Legacy Survey of Space and Time (LSST) will observe over 1,000 supernovae every night – orders of magnitude more than present surveys. The time-domain community needs a method to sift through this volume of data and decide what <1% of objects to follow up spectroscopically in order to maximize the overall science return. However, quantifying the scientific return is complex as the community has diverse science goals, and therefore different criteria for follow-up. During this Incubator we expanded the Recommendation System for Spectroscopic Follow-up (RESSPECT) to incorporate custom user-specified science metrics into either a shared or personal instance.

RESSPECT was originally designed to monitor the Rubin alert stream and select young type Ia supernovae for the Dark Energy Science Collaboration (DESC). Now, there is a modular component to RESSPECT, a user can easily apply their own feature extractor, AI classifier, and value metric to return personalized recommendations for any science case – anomaly detection, multi-messenger astrophysics, follow-up of TNOs – that the user can define as python code. As a test case, anomaly detection has been implemented into RESSPECT using the Lightcurve Anomaly Identification and Similarity Search (LAISS, Aleo et al. 2024) as the AI engine. RESSPECT uses active learning – incorporating the labels from follow-up iteratively – ensuring the framework improves over the duration of LSST. A docker container of RESSPECT has been created and is now deployable on any computer with any resources a user has access to.

Orbit Fitting at LSST Scale

Primary Investigator: Matt Holman, Center for Astrophysics | Harvard & Smithsonian

Fitting orbits (the process of taking the observed on-sky positions and velocities of newly discovered moving Solar System objects and transforming them into orbital parameters) is essential to LSST Solar System science. Discovery and orbital classification of new Solar System small bodies are the top priorities in the LSST Solar System Science Collaboration’s (SSSC’s) Roadmap, but there is no orbit fitting package that can support the needs of the planetary community in the Rubin era. Orbit fitting is essentially the process of minimizing the chi-square or log-likelihood function between the set of observed sky-plane positions and those predicted by a model. We have the engine to power an orbit fitting package, but at present the code can only be used to fit orbits of thousands of objects on a laptop. To handle the LSST era, we need our functions scaled up to handle millions of orbits. This incubator will focus on gracefully parallelizing and scaling up our algorithms for high performance computing to create a highly optimized professional-grade orbit fitting python package. The software is intended to be a resource for the planetary astronomy community. We expect that the orbit-fitting tools developed in this incubator will become a central component of KBMOD [Github] shift-and-stack discovery community software and Rubin Observatory’s Solar System Processing (SSP) data release catalog pipeline.

A Scarlet2 framework for characterizing transients and their host galaxies

Primary Investigator: Charlotte Ward, Penn State University

We are working with the LINCC frameworks team on improving community accessibility to Scarlet2, a GPU-compatible scene modeling code that can model galaxies and transients in multi-resolution, multi-epoch imaging data. We aim for this software to be useful for extracting galaxy SEDs and morphologies from deep coadds, as well as characterizing transients/AGN and their host galaxies after initial discovery by the difference imaging alerts. Scarlet2 is most useful in situations where combining data from Rubin and an overlapping survey will enable improved galaxy models (via incorporating a higher-resolution HST, Euclid or Roman image for the modeling) or improved transient light curves (via incorporating complementary LS4 or ZTF data, for example). Some key applications include: identifying transients as nuclear/non-nuclear (by getting a well-estimated distance from their host galaxy nucleus), extracting transient fluxes when no reference image is available, combining transient photometry across surveys without concern for reference image mismatch, decomposing AGN and host galaxy emission, deep forced photometry where there is uncertainty over the source position, and identifying and modeling host galaxies in crowded fields. With the help of the LINCC team, we look forward to making this code more user-friendly and flexible for the wider Rubin community, so that new users can implement custom Scarlet2 analysis for their science cases.

Time-series Feature Generation and Machine Learning Classification of the Vera C. Rubin Observatory alert
stream

Primary Investigator: Argyro Sasli, University of Minnesota

Linking a Physical Model Library to TDAstro and Improving Performance for Scalable Inference

Primary Investigator: Nikhil Sarin, University of Cambridge