Open Access
Issue
A&A
Volume 667, November 2022
Article Number A141
Number of page(s) 23
Section Catalogs and data
DOI https://doi.org/10.1051/0004-6361/202243745
Published online 18 November 2022

© E. O. Garvin et al. 2022

Licence Creative CommonsOpen Access article, published by EDP Sciences, under the terms of the Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

This article is published in open access under the Subscribe-to-Open model. Subscribe to A&A to support open access publication.

1 Introduction

Strong gravitational lensing is a powerful tool that has multiple applications in astrophysics and cosmology. Strong lenses are used to study the dark matter content and distribution in galaxies (Koopmans et al. 2006; Barnabè et al. 2009; Sonnenfeld et al. 2015; Oldham & Auger 2018) and clusters (Richard et al. 2010; Jauzac et al. 2015; Caminha et al. 2019), to study distant galaxies (Marshall et al. 2007; Dessauges-Zavadsky et al. 2015; Swinbank et al. 2015; Cañameras et al. 2017), and to constrain cosmological parameters such as the Hubble constant and the dark energy equation of state from time delay observations (e.g. Suyu et al. 2010, 2014; Sereno & Paraficz 2014; Courbin et al. 2018; Wong et al. 2020; Millon et al. 2020). With the help of strong gravitational lensing of massive clusters, we have also been able to probe galaxy evolution through Ultraviolet luminosity functions (UV LFs; Atek et al. 2014; Livermore et al. 2017), stellar mass functions (Bhatawdekar et al. 2019; Kikuchihara et al. 2020), and UV slopes (Bhatawdekar & Conselice 2021), well into the epoch of reionisation. Strong gravitational lensing is a rare phenomenon which relies on the chance alignment of a foreground object with a large surface mass density (lens) with a bright background object (source) such as a galaxy, a quasar, or a supernova (Kelly et al. 2015). Recently, other lensed sources have also been reported, such as magnified stars and stellar complexes (Welch et al. 2022; Vanzella et al. 2021). Since the discovery of the first gravitational lens (Walsh et al. 1979) approximately one thousand strong lenses have been confirmed overall.

Finding strong gravitational lenses is a difficult outlier detection problem. To date, numerous automated algorithms have been developed for the detection of strong lenses in large-scale surveys, for example through the identification of arcs and rings in multiband imaging (e.g. Alard 2006; Marshall et al. 2009; Gavazzi et al. 2014; Sonnenfeld et al. 2018) or through the blended signatures of lens and source galaxies in fibre spectroscopy (e.g. Bolton et al. 2008; Brownstein et al. 2012; Holwerda et al. 2015; Shu et al. 2017; Talbot et al. 2021). With the latest developments of artificial intelligence, machine learning algorithms and in particular convolutional neural networks (CNNs) have had an increasing number of applications in astronomy, from galaxy classification (Dieleman et al. 2015; Huertas-Company et al. 2015; Walmsley et al. 2020, 2022) to estimating quantities such as photometric redshifts (Samui & Samui Pal 2017; D’Isanto & Polsterer 2018; Schuldt et al. 2021). Strong lens searches have largely benefited from the use of CNNs (e.g. Bom et al. 2017; Schaefer et al. 2018; Petrillo et al. 2019; Jacobs et al. 2019; Cañameras et al. 2020; Huang et al. 2020). The Bologna strong gravitational lens finding challenge has extensively compared the performance of these methods on simulated data resembling future ground- and space-based imaging surveys such as the Large Synoptic Survey Telescope (LSST) and Euclid (Metcalf et al. 2019). While automated search algorithms are generally highly efficient on real data, they remain affected by false positives. They are typically complemented by visual inspection to increase the sample purity; for instance, only ≃ 1–3% of candidates selected by arc-finders and CNNs are highly probable lenses (Sonnenfeld et al. 2018; Petrillo et al. 2019).

One distributed method that has proved successful in the visual identification of strong lenses is citizen science. The general public has been involved in the classification of astronomical data through projects such as Galaxy Zoo (Lintott et al. 2008). The Space Warps project (Marshall et al. 2016) pioneered the crowdsourced identification of strong gravitational lenses in surveys such as the Canada-France-Hawaii Telescope Legacy Survey (CFHTLS; More et al. 2016) and the Hyper Suprime-Cam Subaru Strategic Program (HSC-SSP; Sonnenfeld et al. 2020). Using the Zooniverse framework, Space Warps displays images on a web-based interface and asks volunteers whether a gravitational lens is present in the images. To assess the completeness and uncertainty of the classifications, Space Warps measures the performance of each user on a training set of simulated lenses, and weights the contributions of individual users based on their skill.

These various lens search methods typically focus on specific lens configurations. On the one hand, galaxy-scale strong lens candidates identified from unresolved fibre spectra have massive foreground elliptical and Einstein radii limited by the fibre aperture sizes (e.g. Bolton et al. 2008; Brownstein et al. 2012). On the other hand, searches using multiband images from wide-field ground-based surveys preferentially select candidates with wider image separations (≥2″ from the lens centre) ensuring robust lens and source deblending, and therefore span galaxy-scale and more massive group-scale and cluster-scale foreground halos (e.g. Belokurov et al. 2009; Sonnenfeld et al. 2018; Cañameras et al. 2020). Moreover, the visual grades assigned by strong lens experts and volunteers show large scatter due to the difficulty in distinguishing strong lenses from rings, spirals, mergers, and other contaminants in seeing-limited images (e.g. More et al. 2016; Rojas et al. 2022).

In this study, we employed a crowdsourced approach to detect strong lenses in archival images from the Hubble Space Telescope (HST) in order to benefit from the higher angular resolutions, and to cover a broader range of lens and source galaxy types, lens potentials, and multiple image configurations. Instead of showing volunteers postage stamps of galaxies, we displayed cutouts of HST images sufficiently large to contain tens of objects. Although the original project was not designed for the detection of lenses but for the detection of asteroids, we asked the volunteers on Hubble Asteroid Hunter1 (HAH) to tag possible lenses on the forum of the project, and the science team inspected all the tagged lenses. Several square degrees of HST imaging had been previously searched for the presence of lenses, for example in the HST All-wavelength Extended Groth strip International Survey (AEGIS; Moustakas et al. 2007), the Galaxy Evolution From Morphology And SEDs (GEMS) survey (More et al. 2011), the Cosmic Evolution Survey (COSMOS; Jackson 2008; Faure et al. 2008; Pourrahmani et al. 2018), and over 7 deg2 of archival observations (Pawase et al. 2014). We extended the systematic morphological selection of strong lenses from the HST archive to a much larger area, without colour, brightness, and redshift cuts, in order to provide a significant and diverse sample of robust lens candidates for future spectroscopic follow-up and detailed modelling.

The structure of this paper is as follows. In Sect. 2, we present the input dataset, and in Sect. 3 we describe the inspection, classification, and light profile fitting. The results and properties of strong lens candidates are given in Sect. 4. We make a comparison with other search methods in Sect. 5. Finally, we summarise in Sect. 6. Throughout the paper we adopt the WMAP Seven-Year Cosmological parameters (Jarosik et al. 2011) with (ΩM, ΩΛ, h) = (0.27,0.73,0.71).

2 Data

We analysed archival HST images from the Advanced Camera for Surveys Wide Field Channel (ACS/WFC), and the Wide Field Camera 3 Ultraviolet and Visible (WFC3/UVIS) and Near-infrared (WFC3/IR) channels. These instruments and detectors have the largest field of views, thus the highest chance of containing serendipitously observed strong lenses. The observations were taken between 30 April 2002 (when the ACS camera was installed) and 24 April 2020 for ACS/WFC, and between 25 July 2009 (when the WFC3 obtained first light) and 24 April 2020 for WFC3/UVIS. This data is presented in Kruk et al. (2022). The WFC3/IR images were uploaded later to the project, after the analysis of the ACS and WFC3/UVIS data was completed, and consisted of observations taken between 25 July 2009 and 1 June 2020.

The analysis is complete with data taken and publicly available in the HST archives up to 1 June 2020. Observations based on general observer (GO) proposals are available in the HST archive 1 yr after they are taken; therefore, the last GO observations analysed were taken before June 2019. HST Snapshot observations are available in the archive immediately after they are acquired, and they have been analysed up to June 2020. One of the authors, citizen scientist Claude Cornen inspected the data released in the HST archives after June 2020 and found 12 new strong lens candidates. These were not included in the main analysis, but are presented separately in Appendix C.

In the Hubble Asteroid Hunter (HAH) project volunteers inspected the single-band HST composite images in PNG format available from the European HST (eHST) archive; the same images are also available in the Mikulski Archive for Space Telescopes (MAST). These images were created from the FITS images by first applying autoscaling, which scales the image linearly from the 0.5% pixel level to the 99.5% pixel level (using autoscale = 99.5, which is the default), and then applying a trigonometric asinh scaling2.

The composite HST images were obtained by stacking individual HST dithered exposures, and processed using the standard HST data processing pipeline by STScI. The exposures were aligned and processed with DrizzlePac3 (Gonzaga et al. 2012) for geometric distortion corrections and cosmic ray removal.

In order to optimise the presentation of the images for the visual inspection by the citizen scientists, we split the HST ACS/WFC and WFC3/UVIS composite images into four equal parts, of sizes 101″ × 101″ and 80″ × 80″, respectively. For the WFC3/IR images we used the full frame of size 123″ × 137″ for the search. We selected all the composite HST images available in the archive based on the following criteria: an exposure time greater than 300 s, and a field of view greater than 7 arcmin2 to exclude sub-frames (only for ACS/WFC and WFC3/UVIS). Occasionally the same field was observed multiple times by HST. We showed all the images available, even when the observations overlapped in the targeted region of the sky.

In total, 145 396 cutouts were inspected, corresponding to 45 784 unique observations. The inspected images were in grey-scale, without colour information. The total area imaged by HST and inspected in this project is about 27 deg2. This includes the HST Cosmic Evolution Survey (COSMOS, Koekemoer et al. 2007; Scoville et al. 2007), as well as the Cosmic Assembly Near-infrared Deep Extragalactic Legacy Survey (CANDELS, Grogin et al. 2011). The estimate of the total area imaged takes into account the overlap between the HST pointings of approximately the same region of the sky with different instruments and filters. These overlapping observations were nonetheless inspected individually by the volunteers. Details of the images inspected are shown in Table 1.

Table 1

HST archival images inspected while searching for strong gravitational lenses in the Hubble Asteroid Hunter project.

3 Method

3.1 Visual identification by citizen scientists

We identified gravitational lenses in the Hubble Asteroid Hunter citizen science project (Kruk et al. 2022), which was launched on 20 June 2019 and ran until August 2020. The primary purpose of the project was to identify serendipitously observed asteroid trails in HST images while the telescope was observing targeted objects proposed by the astronomers. In total, 11482 volunteers inspected the 145 396 cutouts, each image being inspected by ten people. The project also had a dedicated forum called Talk4, where the volunteers could tag cutouts containing interesting objects (with hashtags, #) and discuss them.

In contrast to other citizen science projects that are specifically designed to identify strong lenses (such as Space Warps) or projects that have the strong lens classification as part of their workflows (e.g. Galaxy Zoo), Hubble Asteroid Hunter did not provide a classification option for strong lenses. Instead, we asked the volunteers to specifically tag observations containing strong gravitational lenses on Talk, including individual strong lenses (with #gravitational_lens) and clusters of galaxies with arcs corresponding to strong lensing (with #cluster_lens). For training, we provided several examples of strong gravitational lenses based on previously identified lenses in HST images and targeted lenses, in the Tutorial and in the Field Guide of the project. The science team inspected and assessed all gravitational lenses as the project progressed instead of analysing the classifications when the project was completed. Finally, it is worth mentioning that some of the citizen scientists on the project who tagged lenses participated in Space Warps and, therefore, had prior training and familiarity with the appearance of strong lenses.

3.2 Visual identification by authors

In total, 2354 cutouts were tagged as containing individual strong lenses (with #gravitational_lens). Members of the science team (EOG, SK, CC) inspected all the cutouts, selected 417 unique gravitational lens candidates and added them to our catalogue.

The initial catalogue contained the position of the lens, and the instrument and filter used. We used the ESASky5 portal (Giordano et al. 2018) to retrace the sky coordinates of the lenses in the images. ESASky also reports the objects that appear in previous publications, matching their positions in the HST images with astronomical catalogues and papers from the Strasbourg astronomical Data Center (CDS) Simbad database. This eased our work of establishing which lenses had been previously published. We also double-checked with other databases such as the NASA/IPAC Extragalactic Database (NED). We classified the objects as being ‘targeted’ (in the case where the strong lenses were the target of the observations, judged based on the corresponding HST proposal), previously ‘published’ (whether they appeared in other publications), or ‘unpublished’. A total of 165 strong lenses were the targets of the HST observations, for example by the Sloan Lens ACS Survey (SLACS, Bolton et al. 2008), the CFHTLS-Strong Lensing Legacy Survey (SL2S, More et al. 2012), the Cambridge and Sloan Survey of Wide Arcs in the Sky (CASSOWARY, Stark et al. 2013), or the BOSS Emission-Line Lens Survey (BELLS, Shu et al. 2016). These targeted lenses were excluded from our analysis. The remaining 252 strong lenses were not the primary targets of the HST observations. We searched for these objects on ESASky and NED, and enquired with the 96 Principal Investigators of these HST observations whether they appear in previous publications. We found that 54 had been previously identified and appear in other publications. Hence, we were left with 198 unknown, new lens candidates distributed on the sky according to Fig. 1.

Three members of the science team (EOG, SK, CC) assigned a grade to each lens, based on the morphology, the shape of the source image and the quality and availability of a colour image.

We discussed each object and agreed on a final grade. The criteria used for the classification were inspired by the convention of More et al. (2016), and define the likelihood of each object to be a strong lens. Grade A objects are almost certainly lenses. They are the most promising candidates based on the configuration of the lens, and the morphology in the colour- or grey-scale images. Their identification is based on the presence of an image and counter image, and on a clear colour separation between the lens and image (if colour information is available). Grade B objects are highly probable lenses. These are also high-quality candidates showing probable lensing features, but with typically fewer clear and bright counter-images, and less colour information available than those in grade A. Their confirmation requires spectroscopic follow-up or lens modelling. Grade C objects are maybe lenses. These possible strong lens systems mostly have single arcs without clear counter-images. They are more difficult to distinguish from non-lens contaminants, and some could be tidal tails, compact groups, or weakly lensed arcs.

We further classified the morphology of the source images into ‘Arc’, which is the most common class, ‘Double’ (two images), ‘Triple’ (three images), ‘Quad’ (four images), and ‘Ring’ and ‘Cross’ for the rarest Einstein ring and Einstein cross configurations. Additionally, we classified the morphology of the lenses into ‘elliptical’, ‘disc’, and ‘edge-on disc’.

thumbnail Fig. 1

Map in galactic coordinates showing the sky position of the newly discovered lenses serendipitously observed with HST (red points), over the densities of the parent sample of HST pointings analysed in the HAH project (in tiles of ~215 deg2 each). The sky distribution of the HST observations with strong lens candidates is roughly isotropic, excluding the Galactic plane. The grey dotted line represents the ecliptic.

3.3 Measuring the arc radii

A measurable parameter for the strong lenses in our sample is the arc radius, which can be used as a proxy for the Einstein radius. For the strong lenses made of a single arc, we measured the radius of the circle centred on the brightest pixel of the lensing galaxy and tangent to the arc, using the DS9 and the Aladin circle tool. For the other configurations, we drew a circle, or the best matching ellipse, as a substitute for the outer tangential critical curve in the lens plane and measured its radius.

3.4 Galfit modelling of the lenses

In this subsection, we describe the fitting process using GALFIT (Peng et al. 2002) to determine the parameters of the lenses: the magnitudes, effective radii, axis ratios, and position angles. These parameters are important for future tasks, such as lens modelling and source reconstruction, which are not covered in this work.

We fitted a total of 252 lenses with GALFIT, including the 198 previously unknown systems, and the 54 already identified and published, as discussed above. We used the sky position of the lenses to create cutouts with square sizes of 10″, centred on the objects, from the single-band composite HST images available in eHST. If multiple bands were available, we used the reddest band available, which was F814W in the majority of cases. The band used for fitting is shown in Tables A.1 and A.2.

To ensure that only the lenses are fitted, we masked out the other objects in the cutouts, including the arcs and any other background image, with SExtractor (Bertin & Arnouts 1996). We created images of the HST point spread function (PSF) using TinyTim (Krist 1995), for the band corresponding to each image and the position of the object in the corresponding image, which we used for the fitting. We fitted the lenses with a single Sérsic profile (Sersic 1968) described by the following equation:

(1)

Here, Σ(r) is the surface brightness at radius r, re is the half-light radius, Σe is the effective surface brightness, k is a normalisation coefficient and n is the Sérsic index. As starting parameters, we used a magnitude of 20, an axis ratio of 0.9, and an effective radius of 0.5″.

We used a fixed De Vaucouleurs profile (n = 4) for the majority of the lenses (243 galaxies), which we classified as being elliptical galaxies. We fitted the remaining 23 galaxies which we classified as being discs with an exponential Sérsic index (n = 1). We note that in the case of group-scale lenses, we fitted multiple galaxies which we deemed to be the lenses, hence the total number of galaxies fitted is higher than the number of strong lens candidates. We show the GALFIT models and residuals of the A grade lenses in Figs. B.1 and B.2.

To determine the magnitudes, we calculated the zero points based on the quantities provided in the FITS headers, using6:

(2)

where PHOTFLAM is the inverse sensitivity and represents the scaling factor necessary to transform an instrumental flux in units of electrons per second to a physical flux density, and the PHOTPLAM is the pivot wavelength provided in the FITS header and used to derive the instrumental zero point magnitudes. For consistency, we also measured the magnitudes of the lenses with SExtractor (magAUTO and magISO). We compare the magnitudes estimated with GALFIT and SExtractor in Fig. B.3 and find that the measurements are consistent between the different methods.

4 Results

We present the results for the 198 newly identified lens candidates (hereafter ‘discovered’) in Table A.1, and for the 54 previously published lenses (hereafter ‘rediscovered’) in Table A.2. The lenses in the tables are grouped by grades, and postage stamps of the grade A, grade B, and grade C lenses are shown in Figs. 2, 3, and 4, respectively. In the tables, we provide the IAU name, the RA and Dec coordinates of the candidate, the instrument corresponding to the HST observation in which the lens was found, and the filter used for light fitting. We also provide our measurements of the arc radii rarc, as well as the magnitude, the effective radius re, the axis ratio q, and the position angle PA, all measured using GALFIT. We also list our classifications of the Arc, Double, Ring, and Quad configurations.

While the sample mainly consists of galaxy-galaxy lens configurations, some of the lens galaxies may be part of a cluster, in which case the strong lensing effect can be enhanced by the mass distribution of the cluster. In addition, the presence of a structure along the line of sight, either in front of or behind the main lens plane, will also perturb the light deflection. Firstly, we identify the lens candidates that are associated with known foreground galaxy clusters (e.g. from the Massive Cluster Survey (MACS), Abell, and the Reionisation Lensing Cluster Survey (RELICS) catalogues) using the target name of the HST observations. Secondly, to get a more comprehensive overview of the lens candidates located in possible cluster fields, we also cross-match with the positions of cluster candidates selected (1) from the Sloan Digital Sky Survey Data Release 8 (SDSS DR8) with the red-sequence Matched-filter Probabilistic Percolation (redMaP-Per) algorithm (the SDSS catalogue from Rykoff et al. 2016), and (2) in the footprint of the Dark Energy Spectroscopic Instrument (DESI) legacy imaging surveys based on a photometric redshift clustering analysis (DR8, Zou et al. 2021). Both surveys cover a significant fraction of the line of sight towards our HAH lenses. We use a maximum cross-match radius of 3′ corresponding to ≃1 Mpc at z ~ 0.5. Tables A.1 and A.2 indicate whether each lens matches the position of a cluster, and Table A.3 gives additional information on the cluster names and angular separation. In total, 124 out of the 198 new lens discoveries and 32 out of the 54 rediscoveries are within 3′ of the centroid position of a candidate or confirmed cluster.

Finally, we search for redshifts available for the lens galaxies. We provide either the photometric or spectroscopic redshift, if available, retrieved from NED or from the SDSS-IV DR16 (Ahumada et al. 2020). The last column of Table A.2 lists the reference papers for the rediscovered candidates.

thumbnail Fig. 2

Postage stamps of the 78 grade A HAH lenses identified in this study. The green triangles indicate the 33 rediscovered lenses. The lenses are labelled using the sequence part of their name. The postage stamps are 10″ × 10″. The orientation of the images is north up and east to the left.

thumbnail Fig. 3

Postage stamps of the 89 grade B HAH lenses identified in this study. The green triangles indicate the 15 rediscovered lenses. The lenses are labelled using the sequence part of their name. The postage stamps are 10″ × 10″. The orientation of the images is north up and east to the left.

thumbnail Fig. 4

Postage stamps of the 85 grade C HAH lenses identified in this study. The green triangles indicate the six rediscovered lenses. The lenses are labelled using the sequence part of their name. The postage stamps are 10″ × 10″. The orientation of the images is north up and east to the left.

4.1 Distribution of lens properties

In the previous sections, we described how the gravitational lenses were found and classified as discovered and rediscovered lenses and we explained how we fitted them in order to retrieve the lens parameters. In this subsection we describe the results, namely the distributions of properties such as arc radii, and lens redshifts, magnitudes, and effective radii. The properties are grouped by discovered or rediscovered lenses, and by visual grade.

In Fig. 5, we plot the distribution of photometric and spectroscopic redshifts from SDSS and NED for the lens galaxies for which this information was available. The median redshift of our sample is z = 0.41, and the distribution is skewed to higher redshifts, up to z = 1.3. The discovered and rediscovered lenses show similar distributions, while grade A lenses are, on average, at lower redshifts compared to grade Β or C. The redshift distribution of the lenses in our sample is also broadly consistent with the lens galaxy redshifts from previous ground-based searches (e.g. Bolton et al. 2008; Stark et al. 2013).

Figure 6 shows the distribution of magnitudes over all instruments and filters. We find that the magnitudes of the reported objects are distributed around an average value of m = 20.6. Although our sample contains lenses observed with WFC3/IR, which are typically brighter, our average value is 1.3 magnitudes fainter than the average magnitude m = 19.3 of the sample of lenses found by Pawase et al. (2014) in a previous search in ACS images. The violin plots in Fig. 6 show that the magnitudes of the discovered lenses are fainter than those of the rediscovered lenses (m = 20.8 vs. m = 20.1). The faintest lens found has a measured magnitude of m = 26.4 compared to m = 23.6 in the rediscovered group. Moreover, one-quarter of the rediscovered objects were detected above m = 21, while one-quarter of the new discoveries have a magnitude above m = 21.9. Interestingly, grade C lenses are brighter than grade Β or grade A, showing that we did not assign the grade based on brightness, but on the morphology. It is important to note that the magnitudes are plotted together for both the ACS and WFC3 instruments and some are measured in different filters (see Table A.l for the filters used).

The distribution of the arc radii of the lenses is right skewed, with a median of rarc = 1.58″ and a mean of rarc = 1.94″.

Although it is consistent with the distribution of arc radii presented in Pawase et al. (2014), we see in the left plot in Fig. that our detected lenses include more extreme arc radii than their sample. A few of the newly discovered arc radii range up to = 4.5″, which corresponds to group-scale lenses. In addition, for our newly discovered lenses the median arc radius is rarc = 1.23″ and the smaller arcs we find have a radius of only rarc = 0.33″. This shows a relative improvement in finding smaller separation arcs in comparison to previously reported lenses (in the rediscovered group), for which the median rarc = 1.67″ and the smallest arcs have radii rarc = 0.61″. Furthermore, one-quarter of the rediscovered objects were found to have rarc ≤ 0.88″ while one-quarter of our newly discovered objects have rarc < 0.82″. This observed difference in the median arc radii and their distribution (shown in the central plot in Fig. 7) indicates that citizen science as a strong lens detection method has the potential to reveal arcs with smaller angular separations from the source. This potential deserves to be further investigated and quantified in future studies.

The distributions of lens galaxy effective radii (Fig. 8) show a median re = 0.66″. They are consistent across discovered and rediscovered groups and with the distribution in Pawase et al. (2014).

Finally, the mosaic plot in Fig. 9 shows a visual representation of a contingency table of the grades assigned by members of the science team (E.O.G., S.K., C.C.) to the discovered and rediscovered lenses. We can see that the C grade lenses are larger in proportion than the Β grades and A grades lenses in the discovery group, while the A grades are more frequent than both the B and C grades in the rediscovery group.

Overall, the results we present in this subsection highlight the interest and importance of the contribution of citizen scientists and detection by human eye in order to continuously cross-validate and improve the performance of classical algorithms.

thumbnail Fig. 5

Distributions of redshifts. The left panel shows a histogram of photometric and spectroscopic redshifts retrieved with SDSS and NED. The middle panel uses violin plots to show the respective empirical distributions of discovered and rediscovered lenses, while the right panel shows plots per grade groups. The violin plots were fitted using a Gaussian kernel, with the software R.

thumbnail Fig. 6

Distributions of magnitudes. The left panel shows a histogram distribution of the apparent magnitudes as measured with GALFIT (see Tables A.l and A.2 for description of the filters used). The middle panel shows violin plots of the respective empirical distributions of discovered and rediscovered lenses, while the right panel shows the empirical distributions of magnitudes for each grade group.

thumbnail Fig. 7

Distributions of arc radii. The left panel shows a histogram distribution of the measured arc radii. The middle panel uses violin plots to show the respective empirical distributions of discovered and rediscovered lenses, while the right panel shows the empirical distributions of arc radii per grade groups. The distributions are all skewed to the right.

thumbnail Fig. 8

Distributions of effective radii. The left panel shows a histogram of the effective radii measured with GALFIT. The middle and right panels show the empirical Gaussian kernel distributions of lenses, separated per discovery and per grade, respectively.

thumbnail Fig. 9

Mosaic plot showing the discovery groups and grades in a visual representation of a contingency table. The width of the stacked boxes represents the fraction of each grade in the sample, while the height represents the proportion per discovery group. All the grades were assigned by three members of the science team (EOG, SK, CC) after consensus following the individual evaluations. In the discovery group, there is a higher fraction of C grades than A grades. In the rediscovery group, there is a higher fraction of A grades than both Β and C grades. Hence, the mosaic plot indicates that, overall, our grading scheme is rather conservative towards newly discovered lenses.

4.2 Properties of the high-quality candidates

The morphological search conducted in this paper results in strong lenses with a wide diversity of image configurations and lens potentials; their future spectroscopic follow-up and detailed modelling will enable a range of valuable studies. In Sect. 4.2.1, we illustrate possible applications with the highest quality, newly discovered, grade A candidates in Fig. 2. In Sect. 4.2.2, we describe the lensing configurations and we highlight the properties of a few individual lenses.

4.2.1 Prospects for scientific studies

Strong lenses with distant isolated foreground galaxies are promising objects that will improve our understanding of the total mass-density slope γtot of early-type galaxies. While it is firmly established that γtot is nearly isothermal at low redshift (e.g. Treu & Koopmans 2004; Koopmans et al. 2006; Cappellari et al. 2015), its evolution at z < 1 remains debated, with strong lensing studies suggesting a mild increase from z ~ 1 to z ~ 0 (Koopmans et al. 2006; Bolton et al. 2012; Sonnenfeld et al. 2013; Li et al. 2018), and Jeans dynamical modelling and cosmo-logical simulations favouring nearly constant γtot over this period (e.g. Wang et al. 2019; Derkenne et al. 2021). Measurements are mainly restricted to early-type galaxies at z ≲ 0.6 and need to be extended to higher redshifts in order to further test the galaxy evolution models. Systems with the most distant isolated deflectors (e.g. HAH J083420.3+452506.9 at zspec = 0.65) are very useful in this regard, as is the future spectroscopic follow-up of lenses with zphot > 0.6 (e.g. HAH J005403.4+394712.1, HAH J125709.5+282239.7).

Double source plane lenses are useful for inferring tight constraints on the foreground total mass-density profiles (e.g. Tu et al. 2009). Multiple images covering different angular separations from the lens centre, for instance in HAH J121653.9–121104.2 and HAH J132824.5–313204.6, break parameter degeneracies in the lens models. Together with multiband photometry or stellar kinematic measurements, these lensing configurations enable us to characterise the lens dark-matter distributions. Moreover, when the background sources lie at distinct redshifts, measuring the two Einstein radii can provide valuable and independent constraints on the equation of state of dark energy (Collett et al. 2012). Only a handful of double source plane lenses are known to date (e.g. Gavazzi et al. 2008; Tanaka et al. 2016), and in our sample, HAH J002753.2-753730.0 and HAH J132824.5–313204.6 are good candidates. The former shows two opposite thin blue arcs surrounding a foreground elliptical galaxy, and a broad fuzzy ring that could be a distinct background source. Possible applications to cosmology make these strong lenses well-suited for spectroscopic follow-up.

In addition, systems with isolated lens galaxies and bright extended and structured lensed arcs are typically prioritised to search for the presence of foreground dark matter sub-halos and line of sight halos (e.g. Vegetti & Koopmans 2009; Ritondale et al. 2019). Due to their arc surface brightness distributions, HAH J025659.9–163059.5, HAH J105722.7+580046.5, and HAH J113158.4–195451.5 have the potential to put valuable constraints on foreground mass perturbations.

4.2.2 Strong lensing configurations

While Tables A.1 and A.3 indicate the presence of galaxy clusters along the lines of sight towards several candidates, the image separations and the morphologies in Figs. 2, 3, and 4 show that in fact most have galaxy-galaxy strong lens configurations. In particular, the light deflection is dominated by a single member or line of sight galaxy for 20 out of 21 newly discovered grade A objects with foreground confirmed or candidate clusters. Only HAH J025241.6–150025.5 shows an extended arc and is located within 0.5′ from the centroid position of a cluster candidate listed in Zou et al. (2021). Three grade A objects have rarc > 3″, suggesting a major contribution from an extended group- or cluster-scale mass component to the external convergence and shear; for instance, HAH J093325.2+284348.6 comprises a compact group of three foreground galaxies. These three systems are nonetheless absent from our compilation of cluster candidates selected from shallower ground-based imaging. In addition, Fig. 2 shows a compact group at zspec = 0.1895 towards HAH J234106.5–000007.5, but this system has smaller image separation (rarc ≃ 2.3″), and the associated candidate from Zou et al. (2021) is > 2′ away. Towards the low end of the image separation distribution, several grade A candidates have very compact configurations with subarcsec Einstein radii, as best shown by HAH J042044.0–403607.2 with rarc ≃ 0.4″. This extends the current samples of compact galaxy-scale strong lenses (e.g. SLACS, Bolton et al. 2008), and illustrates the new discovery space that will be unlocked by the Euclid survey with comparable, high-resolution imaging over ≃15 000 deg2.

The sample contains a majority of extended, distorted arcs and rings, which are much less ambiguous than configurations with more compact lensed sources. HAH J001538.4–390435.0, HAH J005403.4+394712.1, and HAH J024329.8–593102.7 are examples of systems with pairs of blue lensed arcs, while HAH J061345.7–562750.4 shows a single but particularly elongated arc, and HAH J033603.9–451223.1 and HAH J100141.8+021424.2 exhibit near-complete Einstein rings. The difficulty in identifying doubles lies in their similarity with compact groups and other non-lens contaminants in the grey-scale HST images seen by volunteers during the first stage of the classification. HAH J100251.8+691959.3 is the only new grade A candidate firmly identified as a blue double-imaged source from the inspection of colour images by the science team. HAH J002348.6–244149.6 also shows compact images, but in a quadruple configuration.

In terms of morphological types, our sample contains interesting classes of foreground lens galaxies. While traditional selection methods are mainly restricted to massive elliptical lens galaxies, the angular resolution reached from space is key to extending to less common lens galaxy types, and to asymmetric or more exotic configurations that are particularly difficult to identify from the ground. Firstly, the new A grade HAH J002348.6–244149.6 comprises a moderately inclined lens spiral galaxy with a bright, blue point-like image in the north and three images distributed evenly along a southern thin arc. This system extends the small number of known spiral deflectors with prominent discs (see also, the SWELLS survey, Treu et al. 2011) and, due to the relative positions of multiple images, its modelling has the potential to tightly constrain the galaxy bulge mass. Secondly, HAH J100251.8+691959.3 and HAH J143811.5+464007.6 show edge-on disc galaxies with rarc ≃ 0.6–0.7″. Thirdly, the flexibility of the crowdsourced classification is demonstrated by the complex structure of the previously published A grade lens HAH J133235.0+503237.3 (Ragozzine et al. 2012). A background galaxy is lensed into a thin elongated arc by a spectacular foreground environment, where an edge-on jellyfish galaxy and nearby ellipticals are embedded within the merging cluster A1758N z = 0.279 (Ebeling & Kalita 2019).

Most candidates with SDSS spectroscopic follow-up have large image separations (e.g. HAH J111337.4+221249.2), which limits the contribution from the lensed sources to the emission collected in the 2″ diameter aperture fibres. This limits the number of candidates with spectroscopic redshift measurements for both the lens and source. We nonetheless inspected the SDSS spectra of all grade A objects, and identified multiple emission lines inconsistent with the lens redshift of 0.6495 for HAH J083420.3+452506.9 (σ* = 269 ± 48 km s−1). Our estimate rarc ≃ 0.6″ suggests that the lens and source galaxies are blended in the SDSS spectrum, and that the background source is a star-forming galaxy with detections of CIV-λ11549, HeII-λ1640, OIII-λ1664, and CIII]-λ1908 emission lines (see Fig. 10). The [OII]λλB727 doublet is also tentatively detected at 9942.6 Å and 9949.47 Å. By fitting the smoothed spectrum, we inferred a probable source redshift of ≈1.667. Higher signal-to-noise ratios are nonetheless needed to confirm this estimate.

thumbnail Fig. 10

Spectrum from the Baryon Oscillation Spectroscopic Survey (BOSS) of the newly discovered grade A candidate HAH J083420.3+452506.9 showing multiple emission lines inconsistent with the spectroscopy of the foreground early-type galaxy of 0.6495. The blue and orange lines show the observed spectrum and the best-fit SDSS model for the central lens galaxy, respectively. The vertical lines (from left to right) give the positions of the CIV-λ11549, HeII-λU640, OIII-λU664, and CIII]-λ11908 emission lines at z ≈ 1.667.

5 Discussion

5.1 Citizen science for lens finding

Previous strong lens searches have relied on dedicated citizen science projects, the most well-known example being the Space Warps project (Marshall et al. 2016), which was also built within the Zooniverse framework. Space Warps has demonstrated that visual inspection by volunteers is an efficient data mining tool for strong lenses. It has been successfully applied to the CFHT Legacy Survey finding 29 promising (and 59 total) new lens candidates (More et al. 2016). More recently, Sonnenfeld et al. (2020) used a dedicated automated lens finding algorithm, YATTALENS (Sonnenfeld et al. 2018) and a citizen science approach on Space Warps to find lenses in the Hyper Suprime-Cam (HSC) survey, finding a total of 14 definite lenses (grade A), 129 probable lenses (grade B), and 581 possible (grade C) lenses. Citizen science approaches have also been implemented for lens modelling (Küng et al. 2015).

Our citizen science project and Space Warps employ different approaches for the inspection of images. Space Warps shows the volunteers postage stamps of objects, generally pre-selected with some criteria (for example, objects within a redshift and mass range), in orderto maximise the chances ofdetecting strong lenses and minimise the required volunteer effort. They also inserted simulated lenses in order to calibrate the response of the volunteers and measure the completeness.

In contrast, in our project we show large fields of view to the volunteers and ask them to tag on Talk if they think the image contains a strong lens. While the Space Warps approach provides a measure of completeness, which is difficult to assess in our case, our project provides an unbiased search for strong lenses (since our images were not pre-selected with some criteria) and is able to find more exotic or interesting lenses. Sonnenfeld et al. (2020) also mention in the Space Warps project that there were candidates flagged in the Talk section that would have otherwise been missed using the main workflow of the project.

Additionally, using large fields of view (corresponding to 80″ × 80″ or larger) allowed us to explore the HST images much more quickly than showing individual postage stamps of galaxies. Finally, this is the first time the HST images have been explored with a citizen science approach, showing the benefit of asking volunteers to inspect space-based images spanning decades in time.

5.2 Comparison to previous HST studies

Previous visual searches for strong gravitational lenses in HST images have been undertaken by Faure et al. (2008) and Jackson (2008) in the 1.64 deg2 COSMOS field, and by Pawase et al. (2014) on 7 deg2 of archival HST I-band images. Additionally, Pourrahmani et al. (2018) trained a convolutional neural network on classifications by Faure et al. (2008) to identify strong lenses in COSMOS.

In the first dedicated exploration of the COSMOS field, Faure et al. (2008) identified 67 low-grade strong galaxy-lens candidates, 20 of which displayed multiple images or obvious arcs. Their searches were limited to massive early-type lenses only with arcs at radii smaller than 5″. Jackson (2008) visually explored all galaxies in COSMOS that are brighter than 25 mag, and found two more certain and one probable lens, in addition to the lenses presented in Faure et al. (2008). Pourrahmani et al. (2018) identified 11 additional candidate strong lenses in COSMOS, using the automated LensFlow machine learning algorithm. Nine of the lenses we identified were reported in previous lens searches in COSMOS (shown in Table A.2). We recovered two more lenses in COSMOS (HAH J100142.8+015448.1, HAH J100222.3+023220.0) that were not previously reported.

In the only other archival search for gravitational lenses in HST images, Pawase et al. (2014) explored 7 deg2 ACS/WFC I-band and WFC3/IR F160W HST images taken until 31 August 2011. They found 40 lens candidates in ACS and 9 in the WFC3/IR images. Even though the images we inspected overlap, we only recovered seven strong lenses (four secure cases) in common between the two studies (see Table A.2). There are 12 cases labelled by Pawase et al. (2014) as secure lenses not recovered in this study. In addition, eight of our grade A lenses (HAH J054707.0-390516.3, HAH J145250.0+580135.3, HAH J045413.1+025733.8, HAH J171314.9+602207.7, HAH J005403.4+394712.1, HAH J084833.4+444431.9, HAH J025241.6-150025.5, HAH J100108.4+024029.9) observed by HST before 31 August 2011 were not recovered by Pawase et al. (2014). This relatively small overlap between the two samples is due to the different approaches: postage stamp inspection by experts (in the case of Pawase et al. 2014) and large cutout inspection by an army of citizen scientists (in our case). It also shows the difficulties in the visual inspection searches of strong lenses, as studies differ in how liberal or conservative they are in the definition of lenses.

5.3 Comparison to automated lens search

Strong lens candidates in this paper were drawn from the largest and most extensive search of 27 deg2 archival HSTimages taken until June 2020. The selection is purely based on morphological criteria, and relies on the systematic inspection of HST images without colour, brightness, or redshift pre-selections. This type of crowdsourced classification is expected to recover the majority of lensing configurations, as long as multiple images are spatially resolved (see, e.g. Sonnenfeld et al. 2020), and Sect. 4 illustrates that candidates indeed cover a wide variety of lens and source galaxy types, image multiplicity, and angular distributions, including exotic lenses.

This approach differs from automated searches in several aspects. Firstly, robots are generally run on wide-area, ground-based imaging and spectroscopic surveys, and focus on massive foreground early-type galaxies due to their high lensing cross-section (e.g. Oguri & Marshall 2010) and to their smooth light distributions and typical rest-frame optical spectra facilitating the identification of signatures from background lensed images. This allows memory limitations to be dealt with but restricts automated classifications to relatively small pre-selected samples (Bolton et al. 2008; Belokurov et al. 2009; Shu et al. 2017; Sonnenfeld et al. 2018). Similarly, despite the significant boost in classification accuracy offered by CNNs, recent lens searches on multiband imaging using machine learning have also focused on the population of elliptical lens galaxies (e.g. Jacobs et al. 2019; Petrillo et al. 2019; Cañameras et al. 2020). Secondly, automated algorithms based on supervised neural networks reach the best classification performance (Metcalf et al. 2019), but they are strongly dependent on the content of the ground truth dataset, and require substantial examples of each morphological class and lens configuration for training. While deep residual networks are able to identify strong lenses with various image separations (e.g. Huang et al. 2020), this poses challenges in developing machine learning pipelines that can reach high completeness over broad ranges of lens halo masses (≃ 1011−1015 M, from isolated galaxies to clusters).

Knabel et al. (2020) compare the citizen science classification of galaxy-scale lenses with automated deep learning and spectroscopic selections over a common footprint covered by the Galaxy and Mass Assembly and Kilo Degree surveys. Interestingly, the independent crowdsourced and machine learning searches result in essentially distinct sets of lens candidates. Despite substantial differences in the parent samples, these findings highlight the complementarity between the two approaches for ground-based searches.

Regarding space-based data, automated lens searches in optical and near-infrared imaging are currently also suffering from limitations. For instance, due to the lack of colour information, the deep learning search conducted in HSTdata by Pourrahmani et al. (2018) is restricted to the most massive lens ellipticals. The minor overlap with our present sample illustrates again that even though machine learning reaches higher performance for specific classification tasks (Metcalf et al. 2019), citizen science projects are crucial to accessing the broad diversity of strong lenses.

5.4 Future prospects

Given the comparable angular resolution and sampling of Euclid and HST images, our results highlight the range of strong lensing studies that will become accessible with the wide Euclid survey. Given the large variety of lens configurations targeted in our archival search, and given the inhomogeneous selection process involving images with different depths and taken in different filters, we cannot robustly measure the sample completeness. It is nonetheless interesting to note that our overall number of strong lens candidates is comparable to the rescaled Euclid forecasts. Collett (2015) predicted that 170000 galaxy-scale strong lenses will be discoverable from the final stack of the Euclid survey, for lens early-type galaxies and various assumptions on the source population and detectability criteria. This corresponds to -300 strong lenses discoverable over our 27 deg2 footprint, while we found ≃240 non-targeted candidates from the ACS and WFC3/UVIS classifications, including a few group- and cluster-scale lenses and more exotic lenses that are not considered by Collett (2015).

This adds further evidence that Euclid will be able to find large samples of >100 000 strong lenses, while covering much broader ranges in lens redshifts and extending robust mass measurements to z > 1 . Strong lens searches in Euclid will likely benefit from joining CNN and crowdsourced classifications, for instance by involving citizens in assembling representative training sets, or cleaning the contaminants from CNN selections in order to increase the sample purity.

6 Conclusion

We presented a sample of 198 new high-quality strong lens candidates identified in 18 yr of HST archival data. We performed a systematic search for non-targeted strong gravitational lenses in the entire HST archive using crowdsourcing to inspect observations taken with the ACS and WFC3 instruments between 2002 and 2020.

We found 252 strong gravitational lens candidates appearing serendipitously in HST observations, after excluding the targeted observations of known lenses. Fifty-four of these candidates had been mentioned in previous studies, as presented in Table A.2. A total of 198 strong lens candidates were not previously reported in the literature. We categorised them into 45 grade A, 74 grade B, and 79 grade C lenses based on their morphology, the shape of the source image, and the quality and availability of an HST colour image. Our selection was purely based on morphological criteria and results in strong lenses with various lens and source galaxy types, and various image multiplicities and angular distributions. This shows the potential of citizen science in accessing the broad diversity of strong lenses, and the complementarity with automated classification algorithms.

The sample contains mostly extended, distorted arcs and rings, which are less ambiguous than compact lensed sources. While 124 (63%) of the new strong lenses are associated with foreground spectroscopically confirmed or candidate galaxy clusters, most of them have galaxy-scale configurations, with the light deflection dominated by a single member or line of sight galaxy. For instance, only three new grade A lenses show a major contribution from the cluster-scale mass component. In addition, while we find that a majority of lenses are ellipticals, ≲10% of the foreground lens galaxies have light profiles consistent with exponential discs, corresponding either to edge-on or to moderately inclined spirals. Finally, a few systems have exotic configurations such as possible double source plane lenses.

In terms of distributions, the newly detected lenses are, on average, 1.3 magnitudes fainter than previous HST searches. The angular separations between multiple images are typically smaller than for lenses found in ground-based data. One-quarter of our newly discovered systems have arc radii ≤0.82″, showing the advantage of combining high-resolution HST imaging with crowdsourcing to select the most compact, galaxy-scale strong lenses for galaxy evolution studies. The redshift distribution of the lens galaxies in our sample is consistent with the lens redshifts from previous ground-based searches.

Since we did not restrict our search to postage stamps of massive elliptical galaxies, as is commonly done in strong lens searches, our study constitutes an unbiased search for lenses with the highest resolution currently possible with HST. This overall sample of 252 strong lenses is a useful benchmark for future lens searches in high-resolution images, such as those with Euclid, James Webb Space Telescope (JWST), or Roman.

This paper shows that crowdsourcing is a robust method of performing visual detection of strong lenses. After receiving proper training and knowledge transfer, the volunteers demonstrated their ability to detect classical lenses and outlier, such as exotic lens configurations. Although inspecting large field-of-view images might lead to a decrease in completeness, this approach has the advantage of considerably speeding up the detection process compared to inspecting postage stamps, where a pre-selection of the targets is often necessary. In addition, providing metadata for the images, such as the coordinates and links to ESASky, enabled the volunteers to perform further analysis and identify the objects in the archives. While using colour images for the inspection might have led to a larger number of lenses being identified, multiple bands were only available for a fraction of the Hubble images in the archives. Nevertheless, the volunteers demonstrated that they were able to identify strong lenses in grey-scale images, which is an important lesson for Euclid, where only a single optical band is available. While it is significantly faster for thousands of citizen scientists to inspect the images than for a handful of professional astronomers, it is important to note that it still took 1 yr of volunteer effort to explore more than 45 000 HST observations. Given the scale of Euclid, using a citizen science approach to explore the entire Euclid dataset of ≃15 000 deg2 is not feasible on its own. Combining artificial intelligence methods and crowdsourcing for the detection of strong lenses, for example through iterative training and validation, has thus a strong potential to produce reliable catalogues on shorter timescales and to detect outliers such as exotic lenses. Our study also demonstrates the potential of crowdsourc-ing as a method to search for strong gravitational lenses in the growing JWST archives, as the high-resolution infrared and multi-wavelength observations will reveal even more lenses with higher redshift and smaller Einstein radii.

In conclusion, this series of Hubble Asteroid Hunter papers reaffirms the importance of crowdsourcing in visually detecting complex objects such as strong gravitational lenses and Solar System objects. It also shows the benefits of exploring large archival datasets spanning decades in time to expand the horizons of future research.

Acknowledgements

We acknowledge the tremendous work made by the citizen scientist volunteers on the Hubble Asteroid Hunter project. Their contributions are individually acknowledged on https://www.zooniverse.org/projects/sandorkruk/hubble-asteroid-hunter/about/results. We thank the anonymous referee for their detailed and insightful comments, as they greatly contributed to enhance the quality of our paper. We thank Stella Seitz and Sherry Suyu for insightful discussions on the sample of strong gravitational lenses. This work has been conducted by EOG as part of the European Space Agency (ESA) student internship. S.K. and R.B. gratefully acknowledge support from the ESA Research Fellowship. This paper is based on observations made with the NASA/ESA Hubble Space Telescope, obtained from the data archive atthe Space Telescope Science Institute. STScI is operated by the Association of Universities for Research in Astronomy, Inc. under NASA contract NAS 5-26555. Based on observations made with the NASA/ESA Hubble Space Telescope, and obtained from the Hubble Legacy Archive, which is a collaboration between the Space Telescope Science Institute (STScI/NASA), the Space Telescope European Coordinating Facility (ST-ECF/ESA) and the Canadian Astronomy Data Centre (CADC/NRC/CSA). The new strong lens candidates (Table A.1) were found in the following HST programmes: 9405, 9414, 9427, 9458, 9476, 9483, 9500, 9722, 9753, 9770, 9822, 9836, 10096, 10134, 10152, 10200, 10207, 10325, 10326, 10334, 10395, 10420, 10491, 10496, 10503, 10504, 10505, 10521, 10523, 10569, 10626, 10635, 10816, 10825, 10861, 10875, 10880, 10881, 10997, 11142, 11588, 11597, 11613, 11697, 11734, 12063, 12064, 12104, 12166, 12195, 12209, 12238, 12253, 12286, 12313, 12319, 12362, 12476, 12477, 12515, 12546, 12549, 12555, 12575, 12591, 12756, 12884, 12898, 12937, 13023, 13024, 13307, 13352, 13364, 13393, 13412, 13442, 13495, 13496, 13514, 13641, 13657, 13695, 13698, 13711, 13750, 13845, 13942, 14096, 14098, 14118, 14165, 14199, 14594, 14662, 14766, 14808, 15063, 15117, 15121, 15183, 15212, 15230, 15275, 15287, 15307, 15320, 15378, 15446, 15495, 15608, 15642, 15644, 15654, 15696, 15843, 16025. We are extremely grateful to the PIs of the HST observations for their positive feedback and responses in including the objects in our catalogue and for the useful conversations. This publication uses data generated via the Zooniverse.org platform, development of which is funded by generous support, including a Global Impact Award from Google, and by a grant from the Alfred P. Sloan Foundation. Funding for the SDSS and SDSS-II has been provided by the Alfred P. Sloan Foundation, the Participating Institutions, the National Science Foundation, the U.S. Department of Energy, the National Aeronautics and Space Administration, the Japanese Monbuka-gakusho, the Max Planck Society, and the Higher Education Funding Council for England. The SDSS Web Site is http://www.sdss.org/. The SDSS is managed by the Astrophysical Research Consortium for the Participating Institutions. The Participating Institutions are the American Museum of Natural History, Astrophysical Institute Potsdam, University of Basel, University of Cambridge, Case Western Reserve University, University of Chicago, Drexel University, Fermilab, the Institute for Advanced Study, the Japan Participation Group, Johns Hopkins University, the Joint Institute for Nuclear Astrophysics, the Kavli Institute for Particle Astrophysics and Cosmology, the Korean Scientist Group, the Chinese Academy of Sciences (LAMOST), Los Alamos National Laboratory, the Max-Planck-Institute for Astronomy (MPIA), the Max-Planck-Institute for Astrophysics (MPA), New Mexico State University, Ohio State University, University of Pittsburgh, University of Portsmouth, Princeton University, the United States Naval Observatory and the University of Washington. This research made use of NASA’s Astrophysics Data System Bibliographic Services. This work made extensive use of Astropy (http://www.astropy.org/), a community-developed core Python package for Astronomy (Price-Whelan et al. 2018) and of the Tool for Operations on Catalogues And Tables (TOPCAT (http://www.star.bris.ac.uk/~mbt/); Taylor 2005). This publication used GALFIT (Peng et al. 2002) and SExtractor (Bertin & Arnouts 1996) for the lens fitting and parameter retrieval, as well as the software tool R (R Core Team 2020) and its packages vioplot (Adler & Kelly 2020) and Hmisc (Harrell Jr & Harrell Jr 2019) for preparation of the tables and data visualisation. This work also made use of Xmatch (Budavari & Lee 2013) and TOPCAT to perform cross-matching of the lenses with nearby clusters.

Appendix A Lists of strong lens candidates

Table A.1

Newly discovered lenses.

Table A.2

Rediscovered lenses identified in this study.

Table A.3

Candidate cluster flags.

Appendix B GALFIT modelling of light profiles of lenses

thumbnail Fig. B.1

Postage stamps showing the lens galaxy fitting with GALFIT for the newly discovered HAH grade A candidates. Three lens candidates are presented per line; each HAH lens is indicated by the sequence part of the name. Each postage stamp is composed of (1) the original image used for the fit using the HST instruments and filters indicated in Tables A.1 and A.2, (2) the fitted model using chi-square minimisation in GALFIT, and (3) the residuals obtained by subtracting the fitted model from the original image. The arcs and background galaxies were masked using SExtractor during the fitting process, and thus only the light of the lens was fitted. The postage stamps show the default orientation of the HST images, which can be different from that shown in Fig. 2.

thumbnail Fig. B.2

continued.

thumbnail Fig. B.3

Verification of the accuracy of the GALFIT magnitude measurements by comparison with the measurements from the Source Extractor software SExtractor. Left: Boxplots representing the distributions of the magnitudes measured with Galf it and the Auto and ISO magnitudes measured with SExtractor. In each boxplot, the middle line represents the median of the distribution. The box itself extends from the lower quartile to the upper quartile, covering the interquartile range. The points that fall outside the whiskers situated on each side of the box are considered extreme values. Right: Comparison between the magnitudes measured with Galf it (abscissa) and the ISO and Auto magnitudes (ordinates) from SExtractor. The grey dotted line is a one-to-one line. These two plots do not show systematic differences in the photometry of the foreground lens galaxies between the different measurements.

Appendix C Additional lens candidates

After the end of the Hubble Asteroid Hunter citizen science project, we found 12 serendipitous non-targeted lens candidates from recent imaging observations released in the HST archives after June 2020 (Table C.1 and Fig. C.1). These candidates were identified by citizen scientist Claude Cornen, who inspected the Daily Data Reports7, while discarding calibrations as well as observations of asteroids, comets, stars or clusters.

thumbnail Fig. C.1

Serendipitous, non−targeted lens candidates identified after the end of the Hubble Asteroid Hunter citizen science project, not included in the main analysis of the paper. The HAH lenses from the mosaics are indicated using only the sequence part of their name.

Table C.1

New strong lens candidates identified after the end of the Hubble Asteroid Hunter citizen science project.

References

  1. Adler, D., & Kelly, S. T. 2020, Vioplot: violin plot, r package version 0.3.7 [Google Scholar]
  2. Ahumada, R., Prieto, C. A., Almeida, A., et al. 2020, ApJS, 249, 3 [Google Scholar]
  3. Alard, C. 2006, ArXiv e-prints, [arXiv:astro-ph/0606757] [Google Scholar]
  4. Atek, H., Richard, J., Kneib, J.-P., et al. 2014, ApJ, 786, 60 [Google Scholar]
  5. Barnabè, M., Czoske, O., Koopmans, L. V. E., et al. 2009, MNRAS, 399, 21 [CrossRef] [Google Scholar]
  6. Belokurov, V., Evans, N. W., Hewett, P. C., et al. 2009, MNRAS, 392, 104 [NASA ADS] [CrossRef] [Google Scholar]
  7. Bertin, E., & Arnouts, S. 1996, A&As, 117, 393 [NASA ADS] [CrossRef] [EDP Sciences] [Google Scholar]
  8. Bettoni, D., Falomo, R., Scarpa, R., et al. 2019, ApJ, 873, L14 [NASA ADS] [CrossRef] [Google Scholar]
  9. Bhatawdekar, R., & Conselice, C. J. 2021, ApJ, 909, 144 [NASA ADS] [CrossRef] [Google Scholar]
  10. Bhatawdekar, R., Conselice, C. J., Margalef-Bentabol, B., & Duncan, K. 2019, MNRAS, 486, 3805 [NASA ADS] [CrossRef] [Google Scholar]
  11. Bolton, A. S., Burles, S., Koopmans, L. V. E., et al. 2008, ApJ, 682, 964 [Google Scholar]
  12. Bolton, A. S., Brownstein, J. R., Kochanek, C. S., et al. 2012, ApJ, 757, 82 [Google Scholar]
  13. Bom, C. R., Makler, M., Albuquerque, M. P., & Brandt, C. H. 2017, A&A, 597, A135 [NASA ADS] [CrossRef] [EDP Sciences] [Google Scholar]
  14. Brownstein, J. R., Bolton, A. S., Schlegel, D. J., et al. 2012, ApJ, 744, 41 [Google Scholar]
  15. Budavari, T., & Lee, M. A. 2013, Xmatch: GPU Enhanced Astronomic Catalog Cross-Matching, Astrophysics Source Code Library, [record ascl:1303.021] [Google Scholar]
  16. Cañameras, R., Nesvadba, N., Kneissl, R., et al. 2017, A&A, 604, A117 [Google Scholar]
  17. Cañameras, R., Schuldt, S., Suyu, S. H., et al. 2020, A&A, 644, A163 [Google Scholar]
  18. Cabanac, R. A., Alard, C., Dantel-Fort, M., et al. 2007, A&A, 461, 813 [Google Scholar]
  19. Caminha, G. B., Rosati, P., Grillo, C., et al. 2019, A&A, 632, A36 [Google Scholar]
  20. Cappellari, M., Romanowsky, A. J., Brodie, J. P., et al. 2015, ApJ, 804, L21 [NASA ADS] [CrossRef] [Google Scholar]
  21. Clowe, D., Markevitch, M., Bradac, M., et al. 2012, ApJ, 758, 128 [NASA ADS] [CrossRef] [Google Scholar]
  22. Collett, T. E. 2015, ApJ, 811, 20 [Google Scholar]
  23. Collett, T. E., Auger, M. W., Belokurov, V., Marshall, P. J., & Hall, A. C. 2012, MNRAS, 424, 2864 [NASA ADS] [CrossRef] [Google Scholar]
  24. Courbin, F., Bonvin, V., Buckley-Geer, E., et al. 2018, A&A, 609, A71 [Google Scholar]
  25. Derkenne, C., McDermid, R. M., Poci, A., et al. 2021, MNRAS, 506, 3691 [NASA ADS] [CrossRef] [Google Scholar]
  26. Desprez, G., Richard, J., Jauzac, M., et al. 2018, MNRAS, 479, 2630 [Google Scholar]
  27. Dessauges-Zavadsky, M., Zamojski, M., Schaerer, D., et al. 2015, A&A, 577, A50 [Google Scholar]
  28. Diego, J. M., Broadhurst, T., Benitez, N., Lim, J., & Lam, D. 2015, MNRAS, 449, 588 [Google Scholar]
  29. Dieleman, S., Willett, K. W., & Dambre, J. 2015, MNRAS, 450, 1441 [Google Scholar]
  30. D’Isanto, A., & Polsterer, K. L. 2018, A&A, 609, A111 [Google Scholar]
  31. Ebeling, H., & Kalita, B. S. 2019, ApJ, 882, 127 [Google Scholar]
  32. Ebeling, H., Ma, C.-J., & Barrett, E. 2014, ApJS, 211, 21 [NASA ADS] [CrossRef] [Google Scholar]
  33. Ebeling, H., Stockmann, M., Richard, J., et al. 2018, ApJ, 852, L7 [Google Scholar]
  34. Ebeling, H., Atek, H., Edge, A. C., et al. 2019, Beyond MACS: A Snapshot Survey of the Most Massive Clusters of Galaxies at z = 0.5-1, HST Proposal [Google Scholar]
  35. Fassnacht, C. D., Moustakas, L. A., Casertano, S., et al. 2004, ApJ, 600, L155 [NASA ADS] [CrossRef] [Google Scholar]
  36. Faure, C., Kneib, J.-P., Covone, G., et al. 2008, ApJS, 176, 19 [NASA ADS] [CrossRef] [Google Scholar]
  37. Fort, B., Le Fevre, O., Hammer, F., & Cailloux, M. 1992, ApJ, 399, L125 [NASA ADS] [CrossRef] [Google Scholar]
  38. Gavazzi, R., Treu, T., Koopmans, L. V. E., et al. 2008, ApJ, 677, 1046 [NASA ADS] [CrossRef] [Google Scholar]
  39. Gavazzi, R., Marshall, P. J., Treu, T., & Sonnenfeld, A. 2014, ApJ, 785, 144 [Google Scholar]
  40. Giordano, F., Racero, E., Norman, H., et al. 2018, Astron. Comput., 24, 97 [NASA ADS] [CrossRef] [Google Scholar]
  41. Gonzaga, S., Hack, W., Fruchter, A., & Mack, J. 2012, The DrizzlePac Handbook (Baltimore: STScI) [Google Scholar]
  42. Grillo, C., Lombardi, M., Rosati, P., et al. 2008, A&A, 486, 45 [NASA ADS] [CrossRef] [EDP Sciences] [Google Scholar]
  43. Grogin, N. A., Kocevski, D. D., Faber, S. M., et al. 2011, ApJS, 197, 35 [Google Scholar]
  44. Guzzo, L., Schuecker, P., Böhringer, H., et al. 2009, A&A, 499, 357 [Google Scholar]
  45. Harrell Jr, F. E., & Harrell Jr, M. F. E. 2019, CRAN2018, 2019, 235 [Google Scholar]
  46. Holwerda, B. W., Baldry, I. K., Alpaslan, M., et al. 2015, MNRAS, 449, 4277 [NASA ADS] [CrossRef] [Google Scholar]
  47. Horesh, A., Maoz, D., Ebeling, H., Seidel, G., & Bartelmann, M. 2010, MNRAS, 406, 1318 [NASA ADS] [Google Scholar]
  48. Huang, X., Storfer, C., Ravi, V., et al. 2020, ApJ, 894, 78 [Google Scholar]
  49. Huertas-Company, M., Gravet, R., Cabrera-Vives, G., et al. 2015, ApJS, 221, 8 [Google Scholar]
  50. Jackson, N. 2008, MNRAS, 389, 1311 [NASA ADS] [CrossRef] [Google Scholar]
  51. Jacobs, C., Collett, T., Glazebrook, K., et al. 2019, ApJS, 243, 17 [Google Scholar]
  52. Jarosik, N., Bennett, C. L., Dunkley, J., et al. 2011, ApJS, 192, 14 [Google Scholar]
  53. Jauzac, M., Richard, J., Jullo, E., et al. 2015, MNRAS, 452, 1437 [Google Scholar]
  54. Kelly, P. L., Rodney, S. A., Treu, T., et al. 2015, Science, 347, 1123 [Google Scholar]
  55. Kikuchihara, S., Ouchi, M., Ono, Y., et al. 2020, ApJ, 893, 60 [NASA ADS] [CrossRef] [Google Scholar]
  56. Knabel, S., Steele, R. L., Holwerda, B. W., et al. 2020, AJ, 160, 223 [NASA ADS] [CrossRef] [Google Scholar]
  57. Küng, R., Saha, P., More, A., et al. 2015, MNRAS, 447, 2170 [CrossRef] [Google Scholar]
  58. Koekemoer, A. M., Aussel, H., Calzetti, D., et al. 2007, ApJS, 172, 196 [Google Scholar]
  59. Koester, B. P., Gladders, M. D., Hennawi, J. F., et al. 2010, ApJ, 723, L73 [NASA ADS] [CrossRef] [Google Scholar]
  60. Koopmans, L. V. E., Treu, T., Bolton, A. S., Burles, S., & Moustakas, L. A. 2006, ApJ, 649, 599 [Google Scholar]
  61. Krist, J. 1995, in Astronomical Society of the Pacific Conference Series, Astronomical Data Analysis Software and Systems IV, eds. R. A. Shaw, H. E. Payne, & J. J. E. Hayes, 77, 349 [NASA ADS] [Google Scholar]
  62. Kruk, S., García Martín, P., Popescu, M., et al. 2022, A&A, 661, A85 (Paper I) [NASA ADS] [CrossRef] [EDP Sciences] [Google Scholar]
  63. Lagattuta, D. J., Fassnacht, C. D., Auger, M. W., et al. 2010, ApJ, 716, 1579 [NASA ADS] [CrossRef] [Google Scholar]
  64. Li, R., Shu, Y., & Wang, J. 2018, MNRAS, 480, 431 [NASA ADS] [CrossRef] [Google Scholar]
  65. Lintott, C. J., Schawinski, K., Slosar, A., et al. 2008, MNRAS, 389, 1179 [Google Scholar]
  66. Livermore, R. C., Finkelstein, S. L., & Lotz, J. M. 2017, ApJ, 835, 113 [Google Scholar]
  67. Marshall, P. J., Hogg, D. W., Moustakas, L. A., et al. 2009, ApJ, 694, 924 [NASA ADS] [CrossRef] [Google Scholar]
  68. Marshall, P. J., Treu, T., Melbourne, J., et al. 2007, ApJ, 671, 1196 [NASA ADS] [CrossRef] [Google Scholar]
  69. Marshall, P. J., Verma, A., More, A., et al. 2016, MNRAS, 455, 1171 [NASA ADS] [CrossRef] [Google Scholar]
  70. Metcalf, R. B., Meneghetti, M., Avestruz, C., et al. 2019, A&A, 625, A119 [Google Scholar]
  71. Millon, M., Galan, A., Courbin, F., et al. 2020, A&A, 639, A101 [Google Scholar]
  72. More, A., Jahnke, K., More, S., et al. 2011, ApJ, 734, 69 [NASA ADS] [CrossRef] [Google Scholar]
  73. More, A., Cabanac, R., More, S., et al. 2012, ApJ, 749, 38 [Google Scholar]
  74. More, A., Verma, A., Marshall, P. J., et al. 2016, MNRAS, 455, 1191 [Google Scholar]
  75. Moustakas, L. A., Marshall, P., Newman, J. A., et al. 2007, ApJ, 660, L31 [NASA ADS] [CrossRef] [Google Scholar]
  76. Morishita, T., Abramson, L. E., Treu, T., et al. 2017, ApJ, 835, 254 [NASA ADS] [CrossRef] [Google Scholar]
  77. Oguri, M., & Marshall, P. J. 2010, MNRAS, 405, 2579 [NASA ADS] [Google Scholar]
  78. Oldham, L. J., & Auger, M. W. 2018, MNRAS, 476, 133 [Google Scholar]
  79. Pawase, R. S., Courbin, F., Faure, C., Kokotanekova, R., & Meylan, G. 2014, MNRAS, 439, 3392 [Google Scholar]
  80. Peng, C. Y., Ho, L. C., Impey, C. D., & Rix, H.-W. 2002, AJ, 124, 266 [Google Scholar]
  81. Petrillo, C. E., Tortora, C., Vernardos, G., et al. 2019, MNRAS, 484, 3879 [Google Scholar]
  82. Pourrahmani, M., Nayyeri, H., & Cooray, A. 2018, ApJ, 856, 68 [Google Scholar]
  83. Price-Whelan, A. M., Sipocz, B., Günther, H., et al. 2018, AJ, 156, 123 [NASA ADS] [CrossRef] [Google Scholar]
  84. R Core Team. 2020, R: A Language and Environment for Statistical Computing (Vienna, Austria: R Foundation for Statistical Computing) [Google Scholar]
  85. Ragozzine, B., Clowe, D., Markevitch, M., Gonzalez, A. H., & Bradac, M. 2012, ApJ, 744, 94 [NASA ADS] [CrossRef] [Google Scholar]
  86. Ratnatunga, K. U., Ostrander, E. J., Griffiths, R. E., & Im, M. 1995, ApJ, 453, L5 [NASA ADS] [Google Scholar]
  87. Repp, A., & Ebeling, H. 2018, MNRAS, 479, 844 [NASA ADS] [Google Scholar]
  88. Richard, J., Smith, G. P., Kneib, J.-P., et al. 2010, MNRAS, 404, 325 [NASA ADS] [Google Scholar]
  89. Richard, J., Patricio, V., Martinez, J., et al. 2015, MNRAS, 446, L16 [Google Scholar]
  90. Ritondale, E., Vegetti, S., Despali, G., et al. 2019, MNRAS, 485, 2179 [Google Scholar]
  91. Rojas, K., Savary, E., Clément, B., et al. 2022, A&A, in press https://doi.org/10.1051/0004-6361/202142119 [Google Scholar]
  92. Rykoff, E. S., Rozo, E., Busha, M. T., et al. 2014, ApJ, 785, 104 [Google Scholar]
  93. Rykoff, E. S., Rozo, E., Busha, M. T., et al. 2016, VizieR Online Data Catalog: J/ApJ/785/104 [Google Scholar]
  94. Samui, S., & Samui Pal, S. 2017, New A, 51, 169 [NASA ADS] [CrossRef] [Google Scholar]
  95. Schaefer, C., Geiger, M., Kuntzer, T., & Kneib, J. P. 2018, A&A, 611, A2 [NASA ADS] [CrossRef] [EDP Sciences] [Google Scholar]
  96. Schirmer, M., Suyu, S., Schrabback, T., et al. 2010, A&A, 514, A60 [NASA ADS] [CrossRef] [EDP Sciences] [Google Scholar]
  97. Schuldt, S., Suyu, S. H., Cañameras, R., et al. 2021, A&A, 651, A55 [Google Scholar]
  98. Scoville, N., Aussel, H., Brusa, M., et al. 2007, ApJS, 172, 1 [Google Scholar]
  99. Sereno, M., & Paraficz, D. 2014, MNRAS, 437, 600 [NASA ADS] [CrossRef] [Google Scholar]
  100. Sersic, J. L. 1968, Atlas de Galaxias Australes (Cordoba, Argentina: Observatori Astronomico) [Google Scholar]
  101. Sharon, K., Bayliss, M. B., Dahle, H., et al. 2020, ApJS, 247, 12 [CrossRef] [Google Scholar]
  102. Shu, Y., Bolton, A. S., Kochanek, C. S., et al. 2016, ApJ, 824, 86 [NASA ADS] [CrossRef] [Google Scholar]
  103. Shu, Y., Brownstein, J. R., Bolton, A. S., et al. 2017, ApJ, 851, 48 [Google Scholar]
  104. Smail, I., Swinbank, A. M., Richard, J., et al. 2007, ApJ, 654, L33 [Google Scholar]
  105. Sonnenfeld, A., Treu, T., Gavazzi, R., et al. 2013, ApJ, 777, 98 [Google Scholar]
  106. Sonnenfeld, A., Treu, T., Marshall, P. J., et al. 2015, ApJ, 800, 94 [Google Scholar]
  107. Sonnenfeld, A., Chan, J. H. H., Shu, Y., et al. 2018, PASJ, 70, S29 [Google Scholar]
  108. Sonnenfeld, A., Verma, A., More, A., et al. 2020, A&A, 642, A148 [Google Scholar]
  109. Stark, D. P., Auger, M., Belokurov, V., et al. 2013, MNRAS, 436, 1040 [Google Scholar]
  110. Suyu, S. H., Marshall, P. J., Auger, M. W., et al. 2010, ApJ, 711, 201 [Google Scholar]
  111. Suyu, S. H., Treu, T., Hilbert, S., et al. 2014, ApJ, 788, L35 [Google Scholar]
  112. Swinbank, A. M., Dye, S., Nightingale, J. W., et al. 2015, ApJ, 806, L17 [Google Scholar]
  113. Talbot, M. S., Brownstein, J. R., Dawson, K. S., Kneib, J.-P., & Bautista, J. 2021, MNRAS, 502, 4617 [Google Scholar]
  114. Tanaka, M., Wong, K. C., More, A., et al. 2016, ApJ, 826, L19 [NASA ADS] [CrossRef] [Google Scholar]
  115. Taylor, M. B. 2005, in Astronomical Society of the Pacific Conference Series, Astronomical Data Analysis Software and Systems XIV, eds. P. Shopbell, M. Britton, & R. Ebert, 347, 29 [NASA ADS] [Google Scholar]
  116. Treu, T., & Koopmans, L. V. E. 2004, ApJ, 611, 739 [Google Scholar]
  117. Treu, T., Dutton, A. A., Auger, M. W., et al. 2011, MNRAS, 417, 1601 [Google Scholar]
  118. Tu, H., Gavazzi, R., Limousin, M., et al. 2009, A&A, 501, 475 [NASA ADS] [CrossRef] [EDP Sciences] [Google Scholar]
  119. Vanzella, E., Caminha, G. B., Rosati, P., et al. 2021, A&A, 646, A57 [EDP Sciences] [Google Scholar]
  120. Vegetti, S., & Koopmans, L. V. E. 2009, MNRAS, 392, 945 [Google Scholar]
  121. Walmsley, M., Smith, L., Lintott, C., et al. 2020, MNRAS, 491, 1554 [Google Scholar]
  122. Walmsley, M., Lintott, C., Géron, T., et al. 2022, MNRAS, 509, 3966 [Google Scholar]
  123. Walsh, D., Carswell, R. F., & Weymann, R. J. 1979, Nature, 279, 381 [Google Scholar]
  124. Wang, Y., Vogelsberger, M., Xu, D., et al. 2019, MNRAS, 490, 5722 [NASA ADS] [CrossRef] [Google Scholar]
  125. Welch, B., Coe, D., Diego, J. M., et al. 2022, Nature, 603, 815 [Google Scholar]
  126. Wong, K. C., Suyu, S. H., Chen, G. C. F., et al. 2020, MNRAS, 498, 1420 [Google Scholar]
  127. Zitrin, A., Fabris, A., Merten, J., et al. 2015, ApJ, 801, 44 [Google Scholar]
  128. Zou, H., Gao, J., Xu, X., et al. 2021, ApJS, 253, 56 [Google Scholar]

2

More details on the parameters used can be found at https://hla.stsci.edu/fitscutcgi_interface.html

All Tables

Table 1

HST archival images inspected while searching for strong gravitational lenses in the Hubble Asteroid Hunter project.

Table A.1

Newly discovered lenses.

Table A.2

Rediscovered lenses identified in this study.

Table A.3

Candidate cluster flags.

Table C.1

New strong lens candidates identified after the end of the Hubble Asteroid Hunter citizen science project.

All Figures

thumbnail Fig. 1

Map in galactic coordinates showing the sky position of the newly discovered lenses serendipitously observed with HST (red points), over the densities of the parent sample of HST pointings analysed in the HAH project (in tiles of ~215 deg2 each). The sky distribution of the HST observations with strong lens candidates is roughly isotropic, excluding the Galactic plane. The grey dotted line represents the ecliptic.

In the text
thumbnail Fig. 2

Postage stamps of the 78 grade A HAH lenses identified in this study. The green triangles indicate the 33 rediscovered lenses. The lenses are labelled using the sequence part of their name. The postage stamps are 10″ × 10″. The orientation of the images is north up and east to the left.

In the text
thumbnail Fig. 3

Postage stamps of the 89 grade B HAH lenses identified in this study. The green triangles indicate the 15 rediscovered lenses. The lenses are labelled using the sequence part of their name. The postage stamps are 10″ × 10″. The orientation of the images is north up and east to the left.

In the text
thumbnail Fig. 4

Postage stamps of the 85 grade C HAH lenses identified in this study. The green triangles indicate the six rediscovered lenses. The lenses are labelled using the sequence part of their name. The postage stamps are 10″ × 10″. The orientation of the images is north up and east to the left.

In the text
thumbnail Fig. 5

Distributions of redshifts. The left panel shows a histogram of photometric and spectroscopic redshifts retrieved with SDSS and NED. The middle panel uses violin plots to show the respective empirical distributions of discovered and rediscovered lenses, while the right panel shows plots per grade groups. The violin plots were fitted using a Gaussian kernel, with the software R.

In the text
thumbnail Fig. 6

Distributions of magnitudes. The left panel shows a histogram distribution of the apparent magnitudes as measured with GALFIT (see Tables A.l and A.2 for description of the filters used). The middle panel shows violin plots of the respective empirical distributions of discovered and rediscovered lenses, while the right panel shows the empirical distributions of magnitudes for each grade group.

In the text
thumbnail Fig. 7

Distributions of arc radii. The left panel shows a histogram distribution of the measured arc radii. The middle panel uses violin plots to show the respective empirical distributions of discovered and rediscovered lenses, while the right panel shows the empirical distributions of arc radii per grade groups. The distributions are all skewed to the right.

In the text
thumbnail Fig. 8

Distributions of effective radii. The left panel shows a histogram of the effective radii measured with GALFIT. The middle and right panels show the empirical Gaussian kernel distributions of lenses, separated per discovery and per grade, respectively.

In the text
thumbnail Fig. 9

Mosaic plot showing the discovery groups and grades in a visual representation of a contingency table. The width of the stacked boxes represents the fraction of each grade in the sample, while the height represents the proportion per discovery group. All the grades were assigned by three members of the science team (EOG, SK, CC) after consensus following the individual evaluations. In the discovery group, there is a higher fraction of C grades than A grades. In the rediscovery group, there is a higher fraction of A grades than both Β and C grades. Hence, the mosaic plot indicates that, overall, our grading scheme is rather conservative towards newly discovered lenses.

In the text
thumbnail Fig. 10

Spectrum from the Baryon Oscillation Spectroscopic Survey (BOSS) of the newly discovered grade A candidate HAH J083420.3+452506.9 showing multiple emission lines inconsistent with the spectroscopy of the foreground early-type galaxy of 0.6495. The blue and orange lines show the observed spectrum and the best-fit SDSS model for the central lens galaxy, respectively. The vertical lines (from left to right) give the positions of the CIV-λ11549, HeII-λU640, OIII-λU664, and CIII]-λ11908 emission lines at z ≈ 1.667.

In the text
thumbnail Fig. B.1

Postage stamps showing the lens galaxy fitting with GALFIT for the newly discovered HAH grade A candidates. Three lens candidates are presented per line; each HAH lens is indicated by the sequence part of the name. Each postage stamp is composed of (1) the original image used for the fit using the HST instruments and filters indicated in Tables A.1 and A.2, (2) the fitted model using chi-square minimisation in GALFIT, and (3) the residuals obtained by subtracting the fitted model from the original image. The arcs and background galaxies were masked using SExtractor during the fitting process, and thus only the light of the lens was fitted. The postage stamps show the default orientation of the HST images, which can be different from that shown in Fig. 2.

In the text
thumbnail Fig. B.2

continued.

In the text
thumbnail Fig. B.3

Verification of the accuracy of the GALFIT magnitude measurements by comparison with the measurements from the Source Extractor software SExtractor. Left: Boxplots representing the distributions of the magnitudes measured with Galf it and the Auto and ISO magnitudes measured with SExtractor. In each boxplot, the middle line represents the median of the distribution. The box itself extends from the lower quartile to the upper quartile, covering the interquartile range. The points that fall outside the whiskers situated on each side of the box are considered extreme values. Right: Comparison between the magnitudes measured with Galf it (abscissa) and the ISO and Auto magnitudes (ordinates) from SExtractor. The grey dotted line is a one-to-one line. These two plots do not show systematic differences in the photometry of the foreground lens galaxies between the different measurements.

In the text
thumbnail Fig. C.1

Serendipitous, non−targeted lens candidates identified after the end of the Hubble Asteroid Hunter citizen science project, not included in the main analysis of the paper. The HAH lenses from the mosaics are indicated using only the sequence part of their name.

In the text

Current usage metrics show cumulative count of Article Views (full-text article views including HTML views, PDF and ePub downloads, according to the available data) and Abstracts Views on Vision4Press platform.

Data correspond to usage on the plateform after 2015. The current usage metrics is available 48-96 hours after online publication and is updated daily on week days.

Initial download of the metrics may take a while.