Search arXiv⌕ Search
Home/Archive/arXiv

arXiv · 2610.10986

DeepDISC on LSST Data Preview 1: Scene-Level Detection, Deblending, and Star/Galaxy Classification in the ECDFS and EDFS Fields

Prince Yadav·Yaswant Sai Ejjagiri·Xin Liu·Grant Merz·the LSST Dark Energy Science Collaboration
View original source ↗Read source document ↗Export metadata ↓

Abstract

We present the first application of DeepDISC, a scene-level deep-learning framework, to real LSST data, using the Data Preview 1 (DP1) coadded images of the Extended Chandra Deep Field-South and the Euclid Deep Field-South fields. In a single pass, DeepDISC jointly detects, segments, and classifies sources as stars or galaxies, rather than treating these tasks as separate steps as in traditional pipelines. We warm-start from a model pretrained on LSST simulations and fine-tune it on DP1 six-band images, using the DP1 catalog's extendedness flag (point-like versus extended) as provisional class labels. The classification part of the network is then fine-tuned alone on a curated sample of stars and galaxies with spectroscopic, Gaia, and photometric identifications, so the final output is a star/galaxy classification. Fine-tuning on real DP1 observations substantially improves source recovery relative to the un-fine-tuned model, raising galaxy and star completeness from 40.2% and 13% to 84.2% and 49.1% under strict class-aware matching; under class-agnostic matching the fine-tuned model reaches 84.9% and 74.3%. On a common sample of sources that neither classifier saw in training, DeepDISC classifies to within about one percentage point of a published photometry-based Random Forest for the same field (97.4% vs. 98.5%); at fainter magnitudes DeepDISC labels more sources as stars than the Random Forest, and independent labels favor the Random Forest in most of these disagreements. Cross-matching to deeper HST/CANDELS imaging on a held-out footprint gives an unrecognized-blend fraction for DeepDISC of 14.9% (95% CI 12.5-17.8%), consistent with or modestly below the 18.0% we measure for the production DP1 catalog. This work establishes a baseline for DeepDISC on real LSST data and a foundation for self-supervised pretraining and for LSST Data Preview 2.

Explore related subjects

astro-ph.IM

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

Prince Yadav, Yaswant Sai Ejjagiri, Xin Liu, Grant Merz, the LSST Dark Energy Science Collaboration. 2026-10-07. DeepDISC on LSST Data Preview 1: Scene-Level Detection, Deblending, and Star/Galaxy Classification in the ECDFS and EDFS Fields. https://arxiv.org/abs/2610.10986

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related papers

arXiv2026

Validation of Satellite Lifetime Predictions at Leonid Space

This paper validates Leonid Space's satellite lifetime prediction pipeline through a backtesting campaign of 1283 deorbited satellites spanning six decades of spaceflight, showing that it accurately recreates their last year of flight. The pipeline combines a custom high-speed orbit propagator, ballistic coefficient estimation from on-orbit TLE data, atmospheric density corrections from data assimilation, custom space weather forecasts, and probabilistic Monte Carlo analysis. Validation proceeds in four steps that progressively remove the benefit of hindsight: starting from a "perfect knowledge" scenario using actual space weather data and ephemerides, we gradually replace historical data with contemporary predictions to arrive at a fully predictive scenario analogous to a deorbit prediction performed for a spacecraft flying today. These forecasts are scored using the continuously ranked probability score and root mean square calibration error, yielding a median one-year prediction accuracy of 2.4 days (0.7%) with perfect knowledge and 35.3 days (9.7%) in the fully predictive scenario. Solar flux forecasts contribute 75% of the variance in our predictions, dominating all other sources of error. To our knowledge, this is the first comprehensive backtesting of long-term satellite lifetime predictions that reports quantitative results for the fully predictive case, and it establishes a new framework for scoring solar cycle forecasts by grounding forecast accuracy in operationally-relevant satellite lifetime predictions.

astro-ph.IM↗
arXiv2026

A Solar Optical Clock

Precision spectroscopy underpins quantum sensing, atomic timekeeping, and the search for terrestrial-mass exoplanets. Finding an Earth-Sun analog requires fractional splitting of spectral features that rivals the precision of the best laboratory spectroscopy. This corresponds to resolution of a few parts in $10^5$ of a typical stellar spectral line, or a Doppler velocity precision of less than 10 cm s$^{-1}$. Only a few astrophysical spectrometers reach this regime, and only by dispersing starlight with large gratings and averaging thousands of lines. Here, we show that comparable precision can be achieved on a single solar line by leveraging techniques used in quantum sensors and optical clocks. A laser is frequency-locked to the 1565 nm solar Fe I line by coupling to a single mode of the solar thermal field and counted against an optical frequency comb referenced to the SI second, thus creating a solar optical clock. Operating at the fundamental photon-noise limit of thermal heterodyne spectroscopy, we achieve centroid precision of one part in $10^9$ of the optical frequency, or 30 cm s$^{-1}$ in radial velocity. We recover the Sun's five-minute acoustic oscillations and the expected Rossiter-McLaughlin signal during a solar eclipse. Combining these techniques from quantum sensing with a thirty-meter-class telescope could enable our demonstrated precision on the nearest Sun-like stars in a few hours. Multiplexing such absolute frequency measurements across lines with different magnetic sensitivity and formation depth would reduce averaging time and help disentangle stellar granulation and activity from the Doppler signature of terrestrial exoplanets, providing new approaches to address some of the most demanding measurements in astrophysical spectroscopy.

astro-ph.IM↗
arXiv2026

Data-driven characterisation of variable source light curves with multimodal representation learning

Surveys such as the Zwicky Transient Facility (ZTF) and the Vera C. Rubin Observatory Legacy Survey of Space and Time (LSST) produce light curves for hundreds of millions of variables. Their characterisation traditionally relies on supervised classifiers with hand-crafted features and curated labels, limiting them to predefined taxonomies and individual surveys. We present a general, survey-agnostic framework that characterises variable sources without labelled training data, producing low-dimensional representations of light curve properties such as amplitude, periodicity, and asymmetry. A multimodal architecture combines two complementary views, a Transformer encoder on raw observations and a convolutional neural network on a generalised Lomb-Scargle periodogram. Each light curve is randomly split into two temporal halves as independent views, and the branches are trained jointly with VICReg to produce embeddings invariant across halves while keeping each dimension informative. We evaluate on 37 006 ZTF sources spanning 14 variable star classes in four broad physical categories. A linear classifier on the frozen 64-dimensional embeddings achieves 77.1% balanced accuracy for classes and 90.5% for broad categories. Combining both branches improves balanced accuracy by up to 14.9 percentage points over the best single modality, confirming their complementarity. Interpretable light curve properties are encoded on distinct dimensions, and within-class analysis reveals physical substructure unresolved by reference labels. Retraining on ZTF g and i bands without modification reproduces the main results, and concatenating multi-band embeddings improves balanced accuracy to 86.1%. The framework needs no labels, phase folding, or survey-specific preprocessing and applies directly to other surveys with different cadences and wavelength coverage.

astro-ph.IM↗