Skip to content

THATCH

Transient Harvester for Archival Training and Classification from Hubble

THATCH is an open-source pipeline for extracting ML-ready transient light curves, image cutouts, and spectra from the Hubble Space Telescope archive. It bridges the domain gap between ground-based surveys (ZTF, Rubin/LSST) and space-based missions (Roman) by providing empirical space-based training data for transient classifiers.

Why THATCH?

The Nancy Grace Roman Space Telescope's High Latitude Time Domain Survey (HLTDS) will discover ~12,000 Type Ia supernovae and thousands of other transients. Current classifiers are trained on ground-based data, but ground-based and space-based observations differ fundamentally in PSF, backgrounds, pixel scale, and filter bandpasses.

HST is the only existing space-based observatory with 30+ years of transient photometry and spectroscopy — making it the ideal bridge between ground-based training data and Roman deployment.

Features

Module Description
thatch.crossmatch Cross-match transient catalogs against the HST archive
thatch.harvest Download calibrated HST images from MAST
thatch.photometry Aperture photometry with HST aperture corrections
thatch.spectra STIS slit and WFC3 grism spectral extraction
thatch.cutouts Image cutout extraction to HDF5
thatch.catalog Unified catalog builder (Parquet/HDF5)
thatch.tracker Pipeline job tracker for batch processing
thatch.data Upload/download from HuggingFace

Data Access

The THATCH dataset is hosted on HuggingFace:

from thatch.data import load_catalog

catalog = load_catalog()  # Downloads from HuggingFace
print(f"{len(catalog)} measurements across {catalog['object'].nunique()} transients")

:fontawesome-solid-database: Browse Dataset on HuggingFace

Quick Start

pip install thatch
from thatch.crossmatch import check_hst_coverage

result = check_hst_coverage({
    "name": "AT2017gfo", "ra": 197.45037, "dec": -23.38148
})
print(f"{result['n_imaging']} imaging + {result['n_spectroscopy']} spectroscopy")

Photometric Accuracy

THATCH achieves 0.02 mag RMS agreement with published HST photometry, validated against Cowperthwaite+2017 and Lyman+2018 measurements of the kilonova AT2017gfo.