2023MNRAS.521.2342O

We introduce a machine learning method for estimating the sensitivity of strong lens observations to dark matter subhaloes in the lens. Our training data include elliptical power-law lenses, Hubble Deep Field sources, external shear, and noise and PSF for the Euclid VIS instrument. We set the concentration of the subhaloes using a v_max-r_max relation. We then estimate the dark matter subhalo sensitivity in 16 000 simulated strong lens observations with depth and resolution resembling Euclid VIS images. We find that with a 3σ detection threshold, 2.35 per cent of pixels inside twice the Einstein radius are sensitive to subhaloes with a mass M_max ≤ 10¹⁰ M_⊙, 0.03 per cent are sensitive to M_max ≤ 10⁹ M_⊙, and the limit of sensitivity is found to be M_max = 10^8.8 ± 0.2^ M_⊙. Using our sensitivity maps and assuming CDM, we estimate that Euclid-like lenses will yield $1.43^{+0.14}_{-0.11}[f_\mathrm{sub}^{-1}]$ detectable subhaloes per lens in the entire sample, but this increases to $35.6^{+0.9}_{-0.9}[f_\mathrm{sub}^{-1}]$ per lens in the most sensitive lenses. Estimates are given in units of the inverse of the substructure mass fraction $f_\mathrm{sub}^{-1}$. Assuming f_sub = 0.01, one in every 70 lenses in general should yield a detection, or one in every ∼ three lenses in the most sensitive sample. From 170 000 new strong lenses detected by Euclid, we expect ∼2500 new subhalo detections. We find that the expected number of detectable subhaloes in warm dark matter models only changes relative to cold dark matter for models which have already been ruled out, i.e. those with half-mode masses M_hm > 10⁸ M_⊙.