Scarlet2: A multi-resolution forward modeling framework for joint imaging data analysis

In this Tech Talk, originally aired October 6, 2025, we are joined by Charlotte Ward for a talk about the widely-adopted Scarlet deblending framework and its latest developments in a talk titled "Scarlet2: A multi-resolution forward modeling framework for joint analysis of imaging data from Rubin and complementary surveys." Scarlet2 was awarded a LINCC Frameworks Incubator grant to great effect, as also described in this presentation.

LINCC Tech Talks, hosted by LINCC Frameworks Project Scientists Colin Orion Chandler (DiRAC Institute, University of Washington) and Mi Dai (University of Pittsburgh), are typically held on the second Thursday of each month. Additional information and the schedule are available here: https://lsstdiscoveryalliance.org/pro.... LINCC is the LSST Interdisciplinary Network for Collaboration and Computing.

Abstract:
We have entered an exciting decade for survey science, with ground-based surveys such as LSST and space-based surveys such as Euclid providing overlapping imaging datasets across the optical and IR. Scarlet2 is a new GPU-compatible software that enables simultaneous forward modeling of ground and space-based imaging at the pixel level. In this approach, a detailed description of the instrument response is combined with super-resolution non-parametric models of galaxies and transients for a flexible forward model that can simultaneously incorporate data from different telescopes. The LINCC team has developed a range of new features to improve usability, including an interactive questionnaire to build custom Scarlet2 code for different science cases, a suite of input and output validation tests, and functions for easier integration with the Rubin Science Platform. Scarlet2 will enable multi-survey analysis for a range of cosmology and time-domain science applications, and has already been applied to tidal disruption events, strongly lensed supernovae, and AGN.