HST2EUCLID: Realistic Euclid Mocks from HST Data

• HST2EUCLID is a Python tool that converts high-resolution HST images into synthetic Euclid-quality observations by matching filter sets, spatial resolution, pixel scale, and noise characteristics.
• The pipeline follows a multi-step process including surface brightness conversion, PSF matching via TinyTim models, linear filter combination, pixel resampling, and realistic sky noise addition.
• It supports strong-lensing analysis and survey forecasting by validating image realism on deep fields and clusters, while highlighting the need for careful calibration for weak-lensing applications.

HST2EUCLID is a Python code for generating realistic mock Euclid observations from panchromatic Hubble Space Telescope imaging. In the Euclid preparation literature, it is presented as a tool that produces Euclid-like images in the $H_{\rm E}$, $J_{\rm E}$, $Y_{\rm E}$, and $I_{\rm E}$ photometric bands by matching Euclid’s filter set, spatial resolution, pixel scale, depths, and noise properties to deep HST data (Collaboration et al., 28 Aug 2025). The code was used to build a simulated database for the 27 galaxy clusters observed in the Cluster Lensing And Supernova survey with Hubble and the Hubble Frontier Fields program, and it was further applied to blank fields, galaxy-scale lenses, and a Euclid Deep Survey preview field (Collaboration et al., 28 Aug 2025). Within the broader Euclid-preparation program, HST2EUCLID is one implementation of “Euclidisation,” the transformation of high-resolution HST observations into synthetic Euclid data products suitable for algorithm validation, lens-modelling forecasts, and pre-survey training workflows (Collaboration et al., 28 Aug 2025, Collaboration et al., 14 Jan 2025).

1. Definition, scope, and observational setting

HST2EUCLID was introduced to exploit the fact that HST images are much deeper and higher-resolution than Euclid, and can therefore serve as a practical empirical basis for realistic Euclid mocks (Collaboration et al., 28 Aug 2025). The intended outputs are Euclid-quality images that retain the structural complexity of real astrophysical scenes, rather than idealized galaxy models. Because the inputs are real observations, the simulated products incorporate, by construction, the observed complexity of galaxy clusters and deep extragalactic fields (Collaboration et al., 28 Aug 2025).

The code is situated in the observational regime of the Euclid survey, for which the relevant design values quoted in the HST2EUCLID description are coverage of $\simeq 15\,000~{\rm deg}^2$, limiting depths of $I_{\rm E}\approx 24.5$ for $10\sigma$ extended sources and $Y_{\rm E}/J_{\rm E}/H_{\rm E}\approx 24$ for $5\sigma$ point sources, and a PSF FWHM of $\simeq 0.13''$ in $J_{\rm E}$0 (Collaboration et al., 28 Aug 2025). On that basis, realistic Euclid mocks are motivated for several tasks: validation of shape-measurement and photo-$J_{\rm E}$1 algorithms, testing of strong- and weak-lensing analyses on cluster fields, forecasting of number counts and size distributions, generation of training sets for deep-learning applications, and planning of spectroscopic follow-up campaigns (Collaboration et al., 28 Aug 2025).

A closely related line of work in the Euclid preparation series examines whether HST images can be used as inputs for weak-lensing image simulations with sufficiently small calibration biases. That study uses a GalSim-based testing environment and directly addresses systematic effects from PSF modelling, interpolation, resampling, and correlated noise (Collaboration et al., 14 Jan 2025). HST2EUCLID should therefore be understood both as a science-enabling image-synthesis framework and as part of a wider methodological effort to determine when HST-derived Euclid-like data are adequate for precision lensing applications.

2. Pipeline architecture and Euclidisation workflow

The HST2EUCLID pipeline is organized into five main steps, denoted $J_{\rm E}$2–$J_{\rm E}$3, and all operations are controlled by a simple YAML or text parameter file (Collaboration et al., 28 Aug 2025). The starting point is a set of calibrated HST FITS images in units of counts in $J_{\rm E}$4, typically spanning the ACS and WFC3 bands needed to reconstruct Euclid photometric channels.

Step $J_{\rm E}$5 converts each HST input image to surface brightness in AB flux density,

$J_{\rm E}$6

with

$J_{\rm E}$7

After this conversion, all HST bands are reprojected to a common astrometric frame and pixel grid using reproject_exact from Astropy/Reproject (Collaboration et al., 28 Aug 2025).

Step $J_{\rm E}$8 performs PSF matching. TinyTim-derived HST PSF models, with FWHM of order $J_{\rm E}$9–$Y_{\rm E}$0 mas, are transformed to Euclid PSFs with convolution kernels $Y_{\rm E}$1 computed via Photutils.create_matching_kernel, such that

$Y_{\rm E}$2

Each HST image is then convolved with the appropriate kernel for the target Euclid band (Collaboration et al., 28 Aug 2025).

Step $Y_{\rm E}$3 constructs Euclid-band images from linear combinations of HST filters. If $Y_{\rm E}$4 are neighbouring HST fluxes and $Y_{\rm E}$5 is the desired Euclid-band flux, then for a basis of SEDs $Y_{\rm E}$6,

$Y_{\rm E}$7

and, writing $Y_{\rm E}$8 and $Y_{\rm E}$9,

$I_{\rm E}$0

In the implementation described for HST2EUCLID, the adopted relations are the following (Collaboration et al., 28 Aug 2025):

Euclid band	HST relation
$I_{\rm E}$1	$I_{\rm E}$2
$I_{\rm E}$3	$I_{\rm E}$4
$I_{\rm E}$5	$I_{\rm E}$6
$I_{\rm E}$7	$I_{\rm E}$8

Step $I_{\rm E}$9 rebins the convolved, combined images to Euclid pixel scales: $\simeq 15\,000~{\rm deg}^2$0 mas px$\simeq 15\,000~{\rm deg}^2$1 in $\simeq 15\,000~{\rm deg}^2$2 and $\simeq 15\,000~{\rm deg}^2$3 mas px$\simeq 15\,000~{\rm deg}^2$4 in $\simeq 15\,000~{\rm deg}^2$5, again using reproject_exact. The images are then converted back from surface brightness to $\simeq 15\,000~{\rm deg}^2$6 with

$\simeq 15\,000~{\rm deg}^2$7

for all bands in the code description (Collaboration et al., 28 Aug 2025).

Step $\simeq 15\,000~{\rm deg}^2$8 adds Euclid-like sky and detector noise. Assuming HST noise is negligible relative to Euclid noise, the background per pixel is chosen so that a source at the limiting magnitude $\simeq 15\,000~{\rm deg}^2$9 attains the desired $I_{\rm E}\approx 24.5$0 in an aperture of $I_{\rm E}\approx 24.5$1 pixels and exposure $I_{\rm E}\approx 24.5$2: $I_{\rm E}\approx 24.5$3 with

$I_{\rm E}\approx 24.5$4

The final image is produced as a Poisson realization of $I_{\rm E}\approx 24.5$5 (Collaboration et al., 28 Aug 2025).

3. Photometric and PSF foundations

The HST2EUCLID procedure combines empirical image transport with explicit photometric-system assumptions. On the HST side, the input images are converted from instrumental count rates using the header-derived quantities $I_{\rm E}\approx 24.5$6 and $I_{\rm E}\approx 24.5$7; on the Euclid side, the target images are expressed in Euclid-band count rates after PSF matching, filter combination, and pixel-scale conversion (Collaboration et al., 28 Aug 2025).

For the near-infrared bands, the broader Euclid photometric-system description provides the formal context for the $I_{\rm E}\approx 24.5$8, $I_{\rm E}\approx 24.5$9, and $10\sigma$0 passbands. In that framework, the total system throughput is

$10\sigma$1

and the quoted AB zero points are $10\sigma$2, $10\sigma$3, and $10\sigma$4 (Collaboration et al., 2022). That paper also reports field-dependent cut-on and cut-off shifts of up to $10\sigma$5–$10\sigma$6 nm across the focal plane (Collaboration et al., 2022). This suggests that the HST2EUCLID configuration, which uses $10\sigma$7 mag in its image-generation recipe, is aimed at internally consistent mock production rather than a full focal-plane-dependent realization of the Euclid NISP photometric system.

PSF treatment is equally central. HST2EUCLID uses TinyTim-derived HST PSF models as the starting point for convolution-kernel construction (Collaboration et al., 28 Aug 2025). In the parallel weak-lensing validation literature, TinyTim PSFs are also examined directly against HST star fields in F606W and F814W, with focus-recovery scatter $10\sigma$8 and $10\sigma$9 for F606W chip 1 and chip 2, versus $Y_{\rm E}/J_{\rm E}/H_{\rm E}\approx 24$0 and $Y_{\rm E}/J_{\rm E}/H_{\rm E}\approx 24$1 for F814W (Collaboration et al., 14 Jan 2025). Residual PSF-model errors at the $Y_{\rm E}/J_{\rm E}/H_{\rm E}\approx 24$2 level in normalized quadrupole and higher moments were found to induce additive biases if not mitigated (Collaboration et al., 14 Jan 2025). For HST2EUCLID, this establishes a systematic backdrop: PSF transport is operationally straightforward, but PSF fidelity matters whenever the output is intended for calibration-grade lensing use.

4. Validation on blank fields and image realism

Validation of HST2EUCLID was performed with six Hubble Frontier Fields parallel blank fields (Collaboration et al., 28 Aug 2025). In these tests, the measured $Y_{\rm E}/J_{\rm E}/H_{\rm E}\approx 24$3 as a function of $Y_{\rm E}/J_{\rm E}/H_{\rm E}\approx 24$4 follows the target limiting-$Y_{\rm E}/J_{\rm E}/H_{\rm E}\approx 24$5 relations, indicating that the background prescription reproduces the designed depth behaviour in the simulated Euclid bands (Collaboration et al., 28 Aug 2025).

The same validation set was used to measure image-size and number-count statistics. For galaxies brighter than $Y_{\rm E}/J_{\rm E}/H_{\rm E}\approx 24$6 mag in $Y_{\rm E}/J_{\rm E}/H_{\rm E}\approx 24$7, the median FWHM is $Y_{\rm E}/J_{\rm E}/H_{\rm E}\approx 24$8, and $Y_{\rm E}/J_{\rm E}/H_{\rm E}\approx 24$9 of the light falls within a radius $5\sigma$0 aperture (Collaboration et al., 28 Aug 2025). The differential galaxy counts per magnitude in $5\sigma$1 agree with HST number counts from Capak (2007) down to $5\sigma$2 at $5\sigma$3, and the cumulative density is $5\sigma$4 (Collaboration et al., 28 Aug 2025). In the application-oriented summary, the number density is given as $5\sigma$5 at $5\sigma$6, and Euclid is reported to resolve $5\sigma$7 of galaxies to $5\sigma$8 (Collaboration et al., 28 Aug 2025).

These tests establish that the pipeline reproduces survey-level observables rather than merely producing visually degraded HST images. A plausible implication is that HST2EUCLID is most valuable when the target diagnostic is tied to detectable-source statistics, morphology at Euclid resolution, or the fidelity of photometric and lensing workflows under realistic observing conditions.

5. Strong-lensing applications and survey forecasting

A principal scientific demonstration of HST2EUCLID is the Euclidisation of the galaxy cluster MACS J0416.1$5\sigma$92403 for strong-lensing analysis (Collaboration et al., 28 Aug 2025). The reference model denoted B23 contains 4 halos, gas, and 213 dPIE galaxies, constrained by 237 spectroscopically confirmed images. When Euclidised data are used together with spectroscopic redshifts for 31 detected images of 12 sources, the resulting EM$\simeq 0.13''$0 model attains $\simeq 0.13''$1, which is $\simeq 0.13''$2 smaller than the B23 value of $\simeq 0.13''$3. In the EM$\simeq 0.13''$4 configuration, where photometric redshifts are used with $\simeq 0.13''$5, the rms increases to $\simeq 0.13''$6 (Collaboration et al., 28 Aug 2025). The same study reports that the total projected mass within the critical curves is recovered to $\simeq 0.13''$7 (Collaboration et al., 28 Aug 2025).

The code was also used to simulate galaxy-scale strong lenses, specifically SDSS J0946+1006 and HE0435$\simeq 0.13''$81223, in regions representative of the Euclid Wide Survey (Collaboration et al., 28 Aug 2025). For Euclid Deep Survey preparation, it generated a simulated $\simeq 0.13''$9 view of the Chandra Deep Field South at $J_{\rm E}$00 s and $J_{\rm E}$01 at $J_{\rm E}$02 (Collaboration et al., 28 Aug 2025). In the abstract-level framing, the methodology is presented as suitable for validating legacy science cases and for training deep-learning techniques before Euclid data become available (Collaboration et al., 28 Aug 2025).

For cluster lensing in particular, the mock data were used to assess whether Euclid-quality imaging alone is sufficient for model construction. The stated conclusion is conditional: complementary photometric or spectroscopic follow-up campaigns are required to measure the redshifts of multiple images and cluster member galaxies (Collaboration et al., 28 Aug 2025). That requirement places HST2EUCLID not only in the category of image simulation software, but also in the planning infrastructure for coordinated survey follow-up.

6. Limitations, systematics, and relation to weak-lensing calibration

The limitations explicitly identified for HST2EUCLID are primarily spectrophotometric and instrumental (Collaboration et al., 28 Aug 2025). The $J_{\rm E}$03 band relies solely on F160W because HST lacks coverage beyond $J_{\rm E}$04, which entails the assumption $J_{\rm E}$05 over the $J_{\rm E}$06 band. Cosmic rays, stray light, and detector cosmetic effects are not modelled, although the code is described as modular and able to ingest custom cosmic-ray masks or a full instrument simulator such as SIXTE. The current PSF modelling uses TinyTim and is identified as a candidate for replacement with end-to-end Euclid PSF simulations such as Mantisale. Additional planned enhancements include lens-light subtraction, sky gradients, and scattered-light features (Collaboration et al., 28 Aug 2025).

The broader weak-lensing Euclidisation study identifies several systematics that are directly relevant whenever HST-derived mocks are used for shear calibration rather than for general image realism (Collaboration et al., 14 Jan 2025). In that work, the difference between direct Euclid-like simulations and Euclidised HST simulations is quantified by

$J_{\rm E}$07

with Euclid-quality calibration targets $J_{\rm E}$08 and $J_{\rm E}$09 (Collaboration et al., 14 Jan 2025). Using native pixel scales, the Lanczos–15 interpolation kernel, noise-correlation correction, and consistent galaxy signal-to-noise ratios between branches reduced residual multiplicative biases to sub-percent levels; finer sampling at $J_{\rm E}$10 reduced $J_{\rm E}$11 further to $J_{\rm E}$12 at $J_{\rm E}$13, while $J_{\rm E}$14 remained $J_{\rm E}$15 (Collaboration et al., 14 Jan 2025). The same analysis found that a $J_{\rm E}$16 uncertainty in the extra Gaussian noise $J_{\rm E}$17 propagates to $J_{\rm E}$18, and that PSF-model residuals can drive additive bias unless neutralised via random rotations or post-deconvolution isotropisation (Collaboration et al., 14 Jan 2025).

Taken together, these results delineate the current status of HST-derived Euclid simulation. HST2EUCLID already supports realistic multi-band survey emulation, strong-lensing feasibility studies, and training-data generation (Collaboration et al., 28 Aug 2025). The companion calibration literature indicates that, for weak-lensing shear calibration, additional care is required in the treatment of correlated noise, interpolation, PSF anisotropy, and galaxy morphology beyond single Sérsic models (Collaboration et al., 14 Jan 2025). This suggests that HST2EUCLID is mature as a Euclid-preparation framework for realistic image synthesis, while calibration-grade lensing deployment remains coupled to the more stringent systematic controls developed in the wider Euclidisation program.