Recovering NEO Streaks from Roman WFI Data for Planetary Defense
| Program ID | 19035 |
|---|---|
| Science Category | Solar System |
| Program Type | Analysis |
| Category | Medium |
| Principal Investigator | Argyro Sasli |
| PI Institution | University of Minnesota |
| Co-Investigators |
|
| Abstract | The Nancy Grace Roman Space Telescope will serendipitously observe thousands of near-Earth objects (NEOs) during its Core Community Surveys (CCS), yet the default calibration pipeline will misclassify these moving-target streaks as cosmic rays and remove them—irretrievably discarding a unique planetary defense dataset. We propose to develop the machine-learning pipeline infrastructure required to detect, extract, and characterize NEO streaks in Roman Wide Field Instrument (WFI) data. Building on proven streak-detection systems deployed on ZTF (ZStreak, Tails), our hybrid approach exploits Roman's up-the-ramp readout to discriminate NEO streaks—which grow across successive resultants—from instantaneous cosmic-ray hits, combined with deep-learning classification trained on synthetic streak injections and transfer learning from ZTF-heritage models. The pipeline will deliver sub-pixel astrometry (~0.011″, ~27× better than NEO Surveyor), multi-band photometry enabling Bus-DeMeo spectral classification down to ~20–30 m objects below Rubin's multi-band threshold, and cross-survey orbit refinement with Rubin/LSST and, once operational, NEO Surveyor. We conservatively expect several hundred to a few thousand NEO streaks across the HLWAS, HLTDS, and GBTDS, spanning the 20–140+ m "city killer" regime. All software, trained models, and a public Roman NEO catalog will be released open-source and deployed on the Roman Research Nexus, with astrometry promptly reported to the Minor Planet Center—transforming the CCS into a powerful planetary defense asset. |