HST imaging of star-forming clumps in 6 GASP ram-pressure stripped galaxies

Exploiting broad- and narrow-band images of the Hubble Space Telescope from the near-UV to I-band rest frame, we study the star-forming clumps of six galaxies of the GASP sample undergoing strong ram pressure stripping. Clumps are detected in H alpha and near-UV, tracing star formation on different timescales. We consider clumps located in galaxy disks and stripped tails and formed in stripped gas but still close to the disk, called extraplanar. We detect 2406 H alpha-selected clumps (1708 in disks, 375 in extraplanar regions, and 323 in tails) and 3745 UV-selected clumps (2021 disk, 825 extraplanar, and 899 tail clumps). Only 15 per cent of star-forming clumps are spatially resolved, meaning that most are smaller than 140 pc. We study the luminosity and size distribution functions (LDFs and SDFs, respectively) and the luminosity-size relation. The average LDF slope is 1.79 +/- 0.09, while the average SDF slope is 3.1 +/- 0.5. The results suggest that the star formation is turbulence-driven and scale-free, as in main-sequence alaxies. All of the clumps, whether they are in the disks or tails, have an enhanced H alpha luminosity at a given size, compared to the clumps in main-sequence galaxies. Indeed, their H alpha luminosity is closer to that of clumps in starburst galaxies, indicating that ram pressure is able to enhance the luminosity. No striking differences are found among disk and tail clumps, suggesting that the different environments in which they are embedded play a minor role in influencing the star formation.


INTRODUCTION
Star formation is the mechanism driving the condensation of atomic gas from galactic to sub-kpc scales down to the dense cores (on pc/sub-pc scales) in which stars eventually form (Section 4 in Kennicutt & Evans 2012) and therefore studying which processes are capable of influencing it is fundamental to our understanding of galaxy formation and evolution.The bridge between the galactic-and core-scale regimes is represented by ≳ 10 pc-scaled star-forming clumps, with masses ≳ 10 4 M ⊙ (Portegies Zwart et al. 2010).Our knowledge about these structures has greatly improved in the last decade thanks to observational surveys of low redshift galaxies with the Hubble Space Telescope (HST ), which is able to achieve the resolution necessary to study their morphology and size properties (LEGUS, Calzetti et al. 2015;DYNAMO, Fisher et al. 2017;LARS, Messa et al. 2019; PHANGS-HST; Lee et al. 2022).Exploiting LEGUS data, many studies find hints that star formation is a turbulence-driven process fragmenting the gas following a scale-free hierarchy (Elmegreen et al. 2014;Gouliermis et al. 2015;Gouliermis et al. 2017).The hierarchical structure is then reflected also in the emerging spatial distribution of stars formed from such gas (Elmegreen & Falgarone 1996;Elmegreen 2006;Grasha et al. 2017).
Moreover, the star formation mechanism can be strongly influenced by the properties of the local medium in which the star forming clumps form, and this leaves an imprint that can be studied using different diagnostics.
Models describing the fragmentation of star-forming regions as a scale-free, turbulence-driven process predict the mass distribution function of these regions to be a power law with slope 2 (Elmegreen 2006).The GASP team corresponding luminosity distribution function (LDF), if derived using a tracer of a narrow age range, is expected to have a similar slope, provided that 1) the Initial Mass Function (IMF) is well sampled and independent of the mass of the initial cloud from which the clumps are formed and 2) the star formation history (SFH) and therefore the stellar age distributions of all clouds are the same (Elmegreen & Falgarone 1996;Elmegreen 2006).
Indeed, from the observational point of view, LDFs of recently formed star-forming regions are known to be well described by a power law, independently of wavelength, with a drop at luminosities fainter than a peak luminosity L peak due to incompleteness (e.g., Hα: Kennicutt et al. 1989;Santoro et al. 2022; UV Cook et al. 2016;Messa et al. 2019;V-band Larsen 2002;Bastian et al. 2007, R-band Whitmore et al. 2014;IR and radio Mascoop et al. 2021).In general, observed slopes are found to be consistent with 2, even though in some cases there are hints of a slope slightly smaller than 2 (Cook et al. 2016: 1.76 ± 0.3;Santoro et al. 2022: 1.73 ± 0.15).Interestingly, some of these works find the value of the slope to be affected by the local environment.Cook et al. (2016) and Santoro et al. (2022) show that the LDF flattens in regions with high star-formation rate (SFR) surface density (Σ SFR ).The same trend is then reflected in the LDF of clumps belonging to different intergalactic environments: Messa et al. (2018) show that the LDF of UV clumps in the spiral arms of the LE-GUS galaxy M51 is flatter than that of clumps in the inter-arm region (with likely lower Σ SFR ).
The size distribution function (SDF) of star-forming clumps is known to be well described by a power law as well (Kennicutt & Hodge 1980;Gusev 2014), with slopes between 2.5 and 4.5, and a flatter distribution for an increasing level of clustering in the clumps.
The luminosity (L Hα typically)-size relation, which many works (Wisnioski et al. 2012;Cosens et al. 2018 and references therein) have shown to be a linear relation in the logarithmic plane, is another proxy of the properties of star formation.The slope and the normalization of the correlation are thought to be related to the geometrical properties of the HII regions ionized by young stars and to the star-formation rate surface density Σ SFR .As shown in Cosens et al. (2018), clumps in starburst-like environments, therefore with a high Σ SFR , are likely to have a higher Hα luminosity at a given size and to follow a flatter distribution.
Jellyfish galaxies are a great laboratory to study star formation in peculiar regimes and environments.Jellyfish galaxies are cluster galaxies undergoing strong ram-pressure stripping (RPS, Gunn & Gott 1972).The ram pressure exerted by the intracluster medium (ICM) is able to strip the gas from the galactic interstellar medium (ISM), eventually producing tails up to more than 100 kpc long, but leaving the stellar disk almost undisturbed (Poggianti et al. 2017a).The gas removal accelerates the quenching of the star formation in the galaxy (Cortese et al. 2021).Several previous works find RPS to be able to briefly enhance star formation during the first stages of the stripping process and to trigger in-situ star formation in compact knots of gas stripped out of the galactic disks (Yoshida et al. 2008;Smith et al. 2010;Merluzzi et al. 2013;Fossati et al. 2016;Consolandi et al. 2017;Jáchym et al. 2019).The tails of these galaxies give the unprecedented opportunity to study the star-formation mechanism in a hotter and higher-pressure environment than the galactic ISM and without the influence of the underlying contamination of the old stellar populations in the disk.Abramson & Kenney (2014) and Kenney et al. (2015) observed galaxies undergoing RPS in the Virgo cluster (NGC 4402 and NGC 4522) and in the Coma cluster (NGC 4921) with HST, respectively.These works show that RPS is able to decouple the high-density component of the ISM (and in particular the Giant Molecular Clouds) from the low-density one, which is more prone to stripping.Also, dust is characterized by elongated morphology and filaments aligned with the stripping direction.Cramer et al. (2019) studied the long and narrow Hα tail of D100 in the Coma cluster with HST and found unbound young UV sources with sizes ∼ 50 − 100 pc, which they consider likely to disperse with ageing.
One of the aims of the GASP (GAs Stripping Phenomena in galaxies with MUSE, Poggianti et al. 2017a) ESO Large Program is to study the properties of galaxies affected by different gas removal processes in the field, in groups and clusters.This includes cluster galaxies in different RPS stages, from pre-stripping (the control sample) to post-stripping (Fritz et al. 2017).Targets were chosen from the catalog in Poggianti et al. (2016) as galaxies with long unilateral debris in optical images, suggestive of gas-only removal.The final sample includes galaxies in the mass range 10 9 − 10 11.5 M ⊙ and at redshift 0.04 < z < 0.07.Targets were observed with MUSE on the VLT (details about the MUSE data are given in Sec.2.2), in order to investigate the properties of both the ionized gas phase and the stellar component in the disks and in the stripped tails.Results of GASP confirmed the presence of clumps with in-situ star formation in the tails of individual jellyfish galaxies of the sample (Bellhouse et al. 2017;Gullieuszik et al. 2017;Moretti et al. 2018;Moretti et al. 2020;Poggianti et al. 2019).In particular, Poggianti et al. (2019) analyzed star-forming clumps in the tails and in the disks of 16 GASP galaxies, finding that tail clumps are less massive, both in terms of stellar and gas mass, and with a smaller gas velocity dispersion than the clumps in the disks.However, the spatial resolution of MUSE at the typical redshift of the GASP galaxies (∼ 1 kpc) did not allow us to study the morphology and the size of these clumps.
In order to better characterize the properties of the star-forming clumps detected in GASP galaxies with MUSE, six galaxies of the GASP sample have been observed with HST, whose resolution is about a factor 14 better than MUSE one (details in Sec.2.1).The broadband filters which were adopted are the F275W, F336W, F606W and F814W, covering a spectral range going from UV-to I-band restframe.In addition, galaxies were observed also with the narrow-band filter F680N, collecting the Hα emission.
This paper is structured as follows: in Sec. 2 we present the HST and MUSE data; in Sec. 3 we define the spatial categories; Sec. 4 is focused on the steps followed for detecting and selecting star-forming clumps and complexes; in Sec. 5 we present the samples of clumps and complexes (5); in Sec.6 we study the luminosity and size distribution functions of clumps and complexes (Secs.6.1 and 6.2, respectively); Sec.7 is dedicated to the luminosity-size relation of the clumps; in Sec. 8 the catalogs which will be publicly available are described; in Sec. 9 we summarize our results.

HST data
In this work, we focus on the study of the luminosities and sizes of star-forming clumps and complexes in a subsample of 6 GASP galaxies, whose main properties are listed in Table 1.The galaxies were selected from the GASP sample of ram-pressure stripped galaxies (Poggianti et al. 2017b) for their extended Hα emitting tails and, in particular, the large number of Hα clumps detected with MUSE observations (Poggianti et al. 2019, see also Vulcani et al. (2020)).
The galaxies were observed using the WFC3/UVIS on board of the Hubble Space Telescope (HST ), using 4 broad-band filters (F275W, F336W, F606W, F814W), which cover a spectral range from UV-to I-band restframe.In addition, galaxies were also observed with a narrow-band filter, F680N, in order to collect the Hα emission at the redshift of these galaxies.Details about observations, data reduction, calibration and analysis, estimate of the standard deviation of the background in each band (σ, hereafter) and Hα + [NII] extraction from the F680N band are described in Gullieuszik et al. (2023), but here we summarise the most important properties.
All images collected have pixel angular size of 0.04 ′′ .The UVIS PSFs in all the 5 filters do not change significantly and have a FWHM of 0.07 ′′1 , corresponding to ∼ 70 pc at the redshifts of the clusters hosting these galaxies (0.0424 − 0.0650, see Table 1).Images were reduced and calibrated using AstroDrizzle 2 .To obtain the Hα + [NII] maps, the continuum emission is modelled by linearly interpolating the emission coming from the broad-band filters F606W and F814W and then subtracted to the F680N images.Contamination from emission lines within the two broad-band filters is expected to have only a small (< 10%) impact on our Hα intensity estimates (see Gullieuszik et al. 2023).
For our purposes, in some cases we are going to work also with denoised versions of these HST images.Denoising is performed using a Python software package called PySAP 3 (Farrens et al. 2020).This algorithm expands the image in Fourier series and we set its parameters in order to remove the high-frequency components, which are typically due to noise.We removed the component with the highest frequency, equal to 2 pixels.This procedure allows us to detect also fainter regions without being dominated by noise, but does not yield reliable sizes.This is the reason for not working with denoised images alone.
Throughout this paper, we work on a smaller squared field-of-view (see Table 2) with respect to the entire HST -WFC3/UVIS images (2.67 ′ × 2.67 ′ ), still sufficient to cover the entire extension of the galaxies and their tails.

DEFINITION OF DISK, EXTRAPLANAR AND TAIL REGIONS
Throughout this work, we are interested in studying the effects of local environment on star formation, and thus we aim at distinguishing star-forming regions originating from stripped gas embedded in the cluster environment from those still in the galaxy disk.
In analogy with what was done for the MUSE observations (Gullieuszik et al. 2020), the starting point to define the stripped tails is the definition of the galaxy stellar disk.As already noted in Gullieuszik et al. (2023) (see also Fig. 1), the high spatial resolution of HST allows us to characterize the galaxy substructures and the stellar disk in more detail than what is possible with MUSE.We used the 2σ contour of the reddest photometric band available (F814W) to draw the most external boundary of the stellar optical disk.2σ values range from 2.14 to 2.67 × 10 −21 erg/s/cm 2 / Å per pixel.We will refer to this contour as the galaxy optical contour (white dashed lines in Fig. 1) and we define as tail the region beyond it.
In the disks of these galaxies there are some regions particularly bright in UV (band F275W), faint in optical (band F814W), elongated and aligned in the same direction of the tails.Therefore, they are likely to be young stellar populations formed in gas already stripped by ram pressure, but still inside the galaxy optical contour because of projection effects or because RPS is at an early stage: we call these regions extraplanar.In order to separate the extraplanar regions from those still in the disk, we visually inspected the RGB images and traced an inner disk contour (red solid lines in Fig. 1), using the UV contours of the images as a guide (green contours in Fig. 1) to separate clumps with an elongated appearance and aligned along the likely stripping direction from those who do not.We define as disk the region within the inner disk contour and as extraplanar the region within the galaxy optical contour but outside the inner disk contour.We point out that this process cannot completely separate undisturbed and stripped gas, since it is done via visual inspection and projection effects may prevent a perfect separation of these two categories of gas.

CLUMPS AND COMPLEXES DETECTION
This section presents the procedure we developed to detect star-forming clumps and to measure their properties.This procedure was applied independently to both the F275W and Hα images (Sec.4.2), in order to trace star formation on different timescales (∼ 200 Myr and ∼ 10 Myr, respectively, Kennicutt 1998;Kennicutt & Evans 2012;Haydon et al. 2020).In addition, a different version of it is applied to the F606W images to fully recover the stellar content in the galaxy tails (Sec.4.3).

Preliminary steps and Astrodendro performance
As a first step, foreground and background sources are masked out.This is done using, when available, the spectroscopic information from MUSE and by visually inspecting the RGB images constructed as described in (Gullieuszik et al. 2023), looking for red elliptical or blue spiral-armed sources, likely to be early-type and spiral galaxies, respectively.
The clump detection is performed using Astrodendro 4 , a software package created to compute dendrograms of observed or simulated Astronomical data, classifying them in a hierarchical tree structure.With this software we are able to detect not only bright clumps, 4 http://www.dendrograms.org/but also sub-clumps inside them.Fig. 2 shows an illustration of 3 possible structures that Astrodendro can generate.
Clumps are defined as local maxima on the image; then the image is analyzed at fainter and fainter flux levels and the clumps grow by including fainter pixels.Eventually, at some point, adjacent clumps might blend together.In this case, those clumps stop growing and are defined as children of a common parent clump; for the following steps, when fainter flux levels are considered, only the parent clump keeps growing.When the flux threshold reaches a given value (see min value in Appendix A), the algorithm stops and the tree structure is built: starting from the clumps at the base of the tree (i.e. the most extended ones), to which a level equal to 0 is assigned.Astrodendro retraces the tree and assigns to the sub-clumps a level equal to the level of their parent clump +1.It also generates a mask to define all pixels corresponding to each clump.
The naming convention used to define the position of the clump in the tree hierarchy is as follows: • trunk : clump with level = 0, regardless of whether it contains sub-clumps or not; • branch: clump with level > 0 and parent of other clumps; • leaf : clump with no children sub-clumps.Notice that, according to this definition, a trunk can be also a leaf.

Observed properties
For all the detected clumps, the following quantities are computed by Astrodendro: • the intensity-weighted mean position of the clump in the plane of the sky, hereafter adopted as clump center; • semi-major and semi-minor axes computed as standard deviations of the flux distribution of the clump in the direction of greatest elongation in the plane of the sky; • the radius r core computed as the geometric mean of the major and minor axes; • the exact area of the clump on the plane of the sky A.
In addition, we computed the following quantities: • the flux densities for all the photometric bands, integrated over the clump area A. The flux uncertainties are computed summing two contributes in quadrature: the background noise, computed as a function of the clump area as described in Gullieuszik et al. (2023), and the Poissonian uncertainty on the source counts converted then into flux considering the conversion factor PHOT-FLAM, the exposure time and the Milky-Way dust attenuation; • Luminosity: calculated from the flux densities using the redshift of the cluster hosting the galaxy (column 8 in Table 1).In order to get Hα luminosities, we compute Hα/(Hα + [NII]) for the Hα clumps detected with MUSE in the galaxies of our sample (Poggianti et al. 2019).The median values obtained for each galaxies are listed in Tab. 1 and used to correct the F680N line flux for the NII emission lines; • r core,corr : PSF-corrected core radius.It is computed by subtracting in quadrature the σ of the PSF (FWHM/2.35≃ 0.03 ′′ , see Sec. 2.1) to r core and it is converted in physical scale according to the redshift of the hosting cluster of each galaxy; • the isophotal radius, defined as • size: defined as 2r core,corr .This choice is supported by the fact that r core,corr is defined by the flux distribution of the clump, therefore it is less sensitive to the flux threshold above which clumps are detected (Wisnioski et al. 2012).Similarly to Wisnioski et al. ( 2012), in Fig. 3 we plot twice the PSF-corrected core radius against the isophotal radius, to show that these two quantities almost follow a 1:1 relation.5

Star-forming clumps
Star-forming clumps are identified in the F275W (UVselected clumps) and Hα (Hα-selected clumps) images running Astrodendro with a flux threshold of 2.5σ on the original images and 2σ on the denoised images6 .The two samples are computed independently, meaning that, in principle, some UV-and Hα-selected clumps may overlap if the same region is bright enough in both the filters.Details about the parameters set for Astrodendro and the methods can be found in Appendix A. Throughout the paper, we use only leaf and trunk clumps (LT sample), unless otherwise stated, to avoid considering the same region too many times.
Astrodendro detected an initial total of 6090 Hα and 6259 UV candidates.To minimize the number of spurious detections we adopted the following procedure, which is schematised in the flow chart shown in Fig. 4.
Firstly, for each of the 5 photometric bands, we flagged a clump as detected if its flux has a signal-to-noise ratio SNR7 higher than 2. We then exclude all clumps that were not detected in at least 3 photometric bands or in both F275W and F680N8 .These criteria yield a reliable detection of clumps, as confirmed by subsequent visual inspection.A total 3611 Hα and 2293 UV spurious detections were removed.As an example, in Fig 5 we show the images in the 5 filters and in Hα of four Hα-selected clump candidates of JO201: the first one (upper left panel) is clearly detected in all images; the second one (upper right panel) does not show UV emission but is detected in three optical filters and in Hα; the third one (lower left panel) is detected only in F680N, F275W and Hα; all these three are therefore confirmed star-forming clumps.The last one (lower right panel) shows emission only in F680N and Hα and was therefore rejected.
Outside the stellar disk, Astrodendro is more prone to detect residual cosmic rays and noise peaks as clump candidates in the tails.Cosmic rays and noise peaks are typically compact and bright, like the star-forming clumps we aim to study.For these reasons, we perform an additional check for clumps in the tails.Both the Hαand UV-selected tail clumps are matched with the corresponding catalogs of Hα clumps detected with MUSE observations and described in Poggianti et al. (2019)  the MUSE spectrum at the corresponding position of the clump is consistent with that of the galaxy, or no redshift can be inferred from the MUSE spectrum, but the clump is detected in all HST filters.To infer the redshift, emission lines such as the [NII] 6548, 6583−Hα and Hβ − [OIII] 4958, 5006 triplets and the [SII] 6716, 6730 doublet are fitted to the MUSE spectra, obtained within a circular aperture as close as possible to that of the clump.73 Hα and 216 UV candidates were rejected after this selection.
Finally, 5 UV-selected trunk clumps in the disks of the JO201, JO204, JO206 and JW100 are removed as their sizes and morphologies are such that they cannot be considered clumps, rather than more likely entire parts of the stellar disks.
For studying sizes, we define a sub-sample (resolved sample) of resolved clumps9 by selecting those objects whose PSF-corrected core radius, r core,corr , exceeds the PSF FWHM (0.07 ′′ ), which corresponds to ∼ 140 pc at the typical redshifts of our targets.
Furthermore, when specified in the following, we removed regions whose emission is powered by an AGN.In order to do that, we used the BPT maps (Baldwin et al. 1981) of the MUSE images of the corresponding galaxies (Poggianti et al. 2017b).Adopting the boundary lines by Kauffmann et al. (2003), Kewley et al. (2001) and Sharp & Bland-Hawthorn (2010), the MUSE spaxels were flagged as star forming, composite, AGN or LINER regions according to the line ratios log([NII]/Hα) and log([OIII]/Hβ) (for the spaxels with S/N> 3 for each line).The HST clumps are flagged as the MUSE spaxels they fall into and in the following we remove those flagged as AGN or LINER when studying the luminosities of the Hα-selected clumps.

Star-forming complexes
UV-and Hα-selected clumps probe the emission coming from or due to stars younger than ∼ 10 8 yr and ∼ 10 7 yr, respectively.The contribution from stellar components older than such timescales can be detected in other optical bands used in this analysis, in order to trace the whole stellar populations formed from the stripped gas in the tails.
Therefore we decided to run Astrodendro also on the F606W filter images, which are deeper than the UV images (see Gullieuszik et al. 2023) and are sensitive to older stellar populations with respect to F275W and Hα.Details of the Astrodendro run on the F606W images are given in Appendix A. Only tail trunk clumps are considered, and we retain only F606W clumps overlapping with at least one pixel of any star-forming clump in either the Hα-selected or the UV-selected samples.In the following, we call star-forming complex the union of a F606W clump and each star-forming clump matched to it.

NUMBER OF CLUMPS: DISK, EXTRAPLANAR AND TAIL CLUMPS
All the clumps are shown in Fig. 6 (the complete figure set is available in the online journal).Here the disk, extraplanar and tail clumps can be seen in different colors, and their hierarchical, tree structure can be appreciated from the color shading.In Fig. 7, we show GASP team zoomed-in examples of Hα-selected clumps in JO201, to illustrate the hierarchical structure and the irregular morphologies of these clumps.
Fig. Set 6. Hα-and UV-selected clumps detected in our sample of galaxies.
The largest clumps are found in the disks of JO175, JO201 and JO206 and in the extraplanar region of JO206.As shown by Fig. 7, large disk clumps (red contours) typically contain several sub-clumps (yellow contours), while extraplanar and especially tail clumps often have only one level.One can also appreciate the effects of RPS on extraplanar clumps, like the filamentary structures in JO206 and JW100, which are particularly bright in UV (lower right panels in Figs.6.4 and 6.6).
In the tails, clumps are often aligned in extended linear or arched structures, suggesting the presence of many sub-tails in each galaxy (as already noticed in Bellhouse et al. 2021 andFranchetto et al. 2020, who found sub-tails in these galaxies from MUSE images).Whether clump and complex properties correlate with distance from the galaxy or along its sub-tails and how they are influenced by the properties of the hosting galaxy are beyond the scope of this paper and will be investigated in future works.
Fig. 8 shows a zoomed-in F606W image of some structures in the tails of JO201, to better appreciate the different spatial distributions of star-forming complexes (dark violet contours), Hα-selected clumps (dark orange) and UV-selected clumps (violet).Typically, an Hα-selected clump has a corresponding UV-selected clump, while the viceversa is not true.Indeed, the number of UV-selected clumps is higher than the number of Hα-selected ones (see below).Furthermore, the corresponding UV-selected clump is generally bigger and almost completely encompasses the Hα-selected clump.Similarly, star-forming complexes contain many UVand Hα-selected clumps, embedded in fainter, optical regions.
The number of star-forming clumps and complexes per galaxy is given in Table 3 and Table 4, respectively.In total, including all galaxies, our LT sample comprises 2406 Hα-selected clumps (1708 disk clumps, 375 extraplanar clumps and 323 tail clumps), 3745 UV-selected clumps (2021 disk clumps, 825 extraplanar clumps and 899 tail clumps) and 424 star-forming complexes.Typically, 98-99% of the selected clumps are leaves (including also simple trunks with no substructures inside), while the trunks containing leaves represent only 1-2% of the whole sample (the fraction increases to 7-14% when restricting the analysis only to resolved clumps).
Avoiding AGN areas and including both resolved and unresolved clumps, ∼ 21% of the Hα-selected and ∼ 7% of the UV-selected clumps get excluded.The percentage is smaller in the latter, indicating that UV-selected clumps are more preferentially located out of AGN regions than Hα-selected clumps.Most of these are disk clumps, as expected, but a few of them can be found in the ionization cone of the AGN, whose extension can reach into the extraplanar region.The exact numbers are listed in brackets in Table 3.
Only 12% of Hα-selected and 16% of UV-selected clumps are spatially resolved, which means that the majority of the clumps have diameters smaller than ∼ 140 pc.Most of the resolved clumps are star-forming according to the BPT, except in the disk where about 25% of the Hα-resolved clumps are flagged as AGN or LINER.
In Fig. 9 we plot the histograms of the number of clumps per galaxy, divided according to the selection band (Hα or UV) and the spatial category (disk, extraplanar and tail), together with the number of complexes.
In most cases disk clumps are much more numerous than extraplanar and tail clumps, regardless of the selection filter, with the only exception being the UV-selected clumps in JW100, which is seen edge-on and is stripped mostly on the plane of the sky (Poggianti et al. 2019), thus in the most favorable conditions to appreciate the extraplanar clumps.For what concerns the number of extraplanar and tail clumps, the prevalence of one over the other depends on the galaxy: in JW100 the number of extraplanar clumps is much larger than that of tail clumps in both the selection filters, in JO204 and JO206 they are almost of the same number, while in the other galaxies tail clumps are more numerous than extraplanar.The number of clumps in each category clearly depends on both the disk inclination and the stripping direction with respect to the line of sight.
Furthermore, with the only exception being disk clumps in JO175, for the same spatial category there are more UV-selected clumps than Hα-selected ones.This indicates that there are a number of stellar-only clumps, with little or no ionized gas left.
The number of star-forming complexes in the tails of the galaxies is generally smaller than the number of tail UV-selected clumps, but larger than that of tail Hαselected ones, with the only exception being JO206, suggesting that many complexes are matched only to UVselected clumps, without any Hα counterpart.

DISTRIBUTION FUNCTIONS
The luminosity (size) distribution function (LDF and SDF hereafter) is defined as the number of sources per luminosity (size) bin normalized by the width of the lu- The complete figure set (6 images) including also the other galaxies is available in the online journal.minosity (size) bin itself and by the total number of sources in the sample and is a useful tool to study the statistical properties of the star-forming clumps.As described is Sec. 1, they are typically well described by a power law, and seem to be good proxies of the environmental effects on the star-formation process and on the clustering properties of the clumps.

Luminosity distribution functions
Figs. 10 and 11 show the histograms of the clumps in each spatial category and in each galaxy, binned in luminosity.The y-axis of plots are normalized by the total number of clumps in the spatial category and in the galaxy.Most Hα-selected clump distributions are peaked at values fainter than ∼ 10 38 erg/s, at the faintend of the luminosity dynamical range.The luminosities of the Hα-selected clumps are consistent with those of "giant" HII regions (like the Carina Nebula), whose Hα luminosities L(Hα) are typically 10 37−39 erg/s, and "super giant" HII regions (like 30 Doradus in the Large Magellanic Cloud), with L(Hα) > 10 39 erg/s (Lee et al. 2011).As expected, the faintest clumps are observed mostly in the closest galaxies of our sample (Table 1).JO201 stands out for its population of bright Hαselected clumps, both in the disk and in the tail, while in the extraplanar regions the brightest clumps are those of JO206 (located in the crest to the top right of the disk, see Fig. 6.4).Similar trends are found for UV-selected clumps.Also, we point out the hint for a bimodality in the distributions of the disk UV-selected clumps of JO201 and JW100 and of the extraplanar clumps of JO206.Finally, the star-forming complexes distributions are very different from the others, since they do not peak in the faint-end of the distribution.
As done in Cook et al. (2016), throughout this work the datapoints of the luminosity distribution functions d N/dL (LDF) are computed fixing the number of objects while varying the bin size, in order to obtain a robust representation of the distribution function.For our LDFs we choose 20 sources per bin.The luminosity GASP team of each bin is the central luminosity of the bin.Datapoints brighter than a given peak luminosity L peak are fitted10 by a power law where K is the normalization and α is the slope of the power law.L peak is chosen for each sub-sample starting from the peak value of the LDF and, if necessary, varying it in order to avoid noisy regions of the LDF.Hα-selected clumps, UV-selected clumps and starforming complexes are fitted independently in each spatial category, in order to study variations in the properties of the LDFs as a consequence of RPS.We used the whole UV-selected and star-forming complexes samples, but only the BPT-selected Hα-selected clumps, in order to avoid AGN-and LINER-powered regions (see Sec. 4.2).Fits were performed using the curve fit method implemented in the SciPy11 Python package, with uncertainties on the LDF computed as the Poisson noise of the number of objects in the bin.
In Figs. 12 and 13 we plot the observed LDFs together with the corresponding best-fitting power laws.Tail LDFs seem to be well described by a single power law, both for Hα-and UV-selected clumps.
In Table 5 we list the best-fitting values of the slopes α and the normalizations K, together with the peak luminosities L peak .Considering all the cases, the value of the slope α is in the range from 1.61 to 1.88 (thus always smaller than 2), with a mean value of 1.79 ± 0.09 (1.84 ± 0.03 for Hα-selected clumps and 1.73 ± 0.09 for UV-selected clumps).In order to rule out the possibility that the inclusion in the sample of trunk clumps with sub-clumps can bias the results, we performed the same fits to the LDFs excluding them.Since this kind of trunks is ∼ 2% of the whole sample, excluding them does not affect significantly the results and the leaf-only slopes are always consistent within 1σ with those obtained including both trunks and leaves.
Our LDF slopes are consistent with previous results for HII regions (Kennicutt et al. 1989: 2.0 ± 0.5) and are smaller than 2. Similar results have been found by Santoro et al. (2022) for HII regions and Cook et al. (2016) for UV-selected young stellar clusters.
We also performed a KS -test on the luminosity distributions for pairs of spatial categories, for Hα-and UVselected clumps separately, to infer whether the LDF significantly changes from one region to another.We compare the distributions above the maximum L peak value above which we can assume all three sub-samples to be complete.The resulting P values are listed in the Appendix (Table A2) and are consistent with what one would expect when comparing the slopes of the LDFs.For Hα-selected clumps, where the slopes are consistent with each other within the errors, the KS-test cannot exclude that each pair of distributions are identical.For UV-selected clumps, the KS-test confirms significant differences for the pairs disk-extraplanar and disk-tail.
In the top panel of Fig. 14 we show the comparison among the best-fitting slopes, as a function of the selection band and the spatial category.Both the UV and Hα slopes steepen going from disk, to extraplanar to tail regions, where the closest match with the expected slope α = 2 is found.
Shallower LDFs are found in galaxies with high sSFR (Santoro et al. 2022), such as all our jellyfish galaxies (Vulcani et al. 2018), which may explain why our slopes are smaller than 2. Furthermore, as described in Sec. 1, past works (Cook et al. 2016;Messa et al. 2018;Santoro et al. 2022) find flatter LDFs in environments with a high SFR surface density Σ SFR .Whether tails and disks are characterized by different Σ SFR is a matter of future works, where masses and SFR od the clumps will GASP team   4) and ( 5) contain the values of the best-fitting slopes α, the best-fitting normalization KL and the peak luminosity L peak arbitrarily chosen, over which datapoints are fitted (Eq.2).Notice that L peak is in erg/s for Hα-selected clumps, whereas it is in erg/s/ Å for UV-selected clumps and star-forming complexes.Columns ( 6), ( 7) and ( 8) list the same quantities (best-fitting slope αs, best-fitting normalization Ks and peak size size peak ), but for the SDFs.are also found in simulations that include the ageing effect of the most massive clumps (Gieles 2009;Fujii & Portegies Zwart 2015), which would be consistent with the fact that the slopes of the Hα-selected clumps (circles in Fig. 14) are larger than those of the UV-selected clumps (squares) of the corresponding spatial category.
It would be also confirmed by the slope of the starforming complexes (all of which are located in the tails by construction), which is very close to that of tail UVselected clumps.
Our analysis therefore suggests that the tails contain proportionally fainter clumps than the disks, and the extraplanar regions are intermediate between the two.However, this difference is statistically significant only when comparing UV-selected disk clumps with the other spatial categories, while for Hα-selected clumps there are only hints of such trend (Fig. 14).Furthermore, observational biases could explain the shallower LDF observed in disk clumps, since disk clumps are expected to be more affected by blending effects and underlying disk contamination, while the tail clumps are the least contaminated population, being isolated.Hence their observed LDF should be the closest to the intrinsic one.Indeed, it is the closest to the theoretical expected value of 2 (Elmegreen 2006).Thus, we can conclude that the properties of the gas in which clumps are embed-  ded are likely to play a minor role in influencing the LDF.Nonetheless, this analysis cannot fully exclude ef-fects on other properties of the clumps, like the mass, which we will investigate in future works.

Deviation from single power law
Carefully inspecting Fig. 12, it is evident that disk (and to some extent, also extraplanar) LDFs show some particular features, such as slope changes, plateaus and secondary peaks, hinting to the need of a more complex model rather than a single power law.To characterize these different regimes, disk LDFs are divided in three intervals: the faint-end interval, the plateau and the bright-end interval, each fitted with a powerlaw.Furthermore, for the Hα-selected LDF we fitted a power law also to datapoints brighter than 1.2 × 10 39 erg/s, in correspondence of a secondary peak of the  5).Circles are Hαselected clumps, squares are UV-selected clumps and triangles are star-forming complexes.Colors refers to the spatial category: red for disk, blue for extraplanar and green for tail.
LDF 12 ("secondary-peak interval", hereafter).We superimposed the best-fitting disk power laws to the extraplanar LDFs, in order to understand if also this spatial category could be characterized by the same regimes (we do not have enough statistics to divide the extraplanar LDFs in intervals and fit a power law in each of them).
12 This secondary peak is dominated by clumps in JO201 (the galaxy with the largest amount of disk and tail clumps).Nonetheless, we do not have reasons to think there is a bias in luminosity artificially increasing the number of clumps at such luminosity, therefore it is a matter of interest to characterize this interval, too.
The best-fitting slopes and the luminosity range boundaries of each interval are shown in Table 6 and in Fig. 15 we show the best-fitting power laws superimposed onto the disk and extraplanar LDFs.
For what concerns Hα-selected LDFs, the faint-end interval slope is larger than that of the bright-end interval, hinting to a change in the properties of the clumps before and after the plateau.When considering the secondarypeak interval, the distribution gets steeper than for the bright-end interval, but still flatter than at the faintend.When superimposing these results on the extraplanar LDF (right-end panels in Fig. 15), we can notice that the faint-and bright-end best-fitting power laws describe quite well the distribution.On the other hand, the extraplanar LDF seems to lack the plateau and the secondary-peak, even though we do not have enough datapoints in these intervals to exclude this hypothesis.
Concerning the disk UV-selected LDF, the slopes in the faint-and bright-end intervals are consistent within the uncertainties.The plateau covers a wider luminosity range compared to the Hα plateau.The presence of a plateau in UV LDFs has never been observed before.Furthermore, the extraplanar LDF is well described by the results obtained for the disk, especially in the faintend interval.
Whether these different regimes are an effect of the ageing or not is not clear yet.The position of the plateau in the disk Hα LDF is compatible with a change in the bounding regime (from density bound to ionization bound ) of the HII regions (Beckman et al. 2000) at a predicted Hα luminosity (at ∼ 4 × 10 38 erg/s).On the other hand the slopes at the low and high luminosity ends are similar, while the Beckman et al. (2000) model predicts a steepening at bright luminosities, where HII regions are ionization bound.Moreover, our LDFs show the same plateau also in the disk UV-selected clumps, which should not be affected by the changing in the ionization regime.

Size distribution functions
In this section we use the clumps of the resolved sample(s).The analysis of the size distribution functions of the clumps (SDFs hereafter) is performed in the same way described in Sect.6.1.The samples are binned using 15 sources per bin for disk clumps and 5 sources per bin for the extraplanar and tail clumps, because of the low number of clumps in these spatial categories.SDFs are qualitatively similar to LDFs.Their intrinsic functional form is a power law, but incompleteness effects introduce a cutoff at small sizes.In analogy with what   we did for the LDF (Eq.2), we define the peak value as size peak and we fit a power law to datapoints above this value.In Fig. 16 the observed SDFs and the best-fitting model of each sub-sample are shown.For completeness, we plot also the SDF datapoints of unresolved clumps, for which we have only upper limits for the sizes (filled dots).In order to do that, SDFs are not normalized for the total number of clumps, since the normalization changes when considering unresolved clumps or not.A single power law is likely to be a good representation of the resolved data, especially considering that the sample is about 15% of the one used to constrain the parameters of the LDFs (see Table 3).The loss of statistics can affect especially the extraplanar and tail sub-samples, for which the regime in which the sample is complete includes just a few datapoints.The fitted power laws do not seem to well describe the unresolved datapoints, as expected from incompleteness.These features, together with the fact that unresolved clumps have, by definition, no reliable estimates of their sizes, imply that we cannot draw any conclusion for sizes below ∼ 140 pc.
The best-fitting slopes and normalizations, and the chosen size peak of each sub-sample are listed in columns ( 6), ( 7) and ( 8) of Table 5.The average slope is 3.3±0.6(3.6 ± 0.6 for Hα-resolved clumps and 3.1 ± 0.3 for UVresolved clumps).Slopes of extraplanar and tail Hαresolved clumps are consistent with the one found by Kennicutt & Hodge (1980) in the disk of a low-z spiral galaxy (α = 4.1).
As done in Sect.6.1 for LDFs, we computed the P values from the KS statistics comparing the size distributions of pairs of spatial categories, keeping the two selection filters separated.Results are listed in Appendix (Table A2).In this case, the KS finds significantly different distributions for all pairs, except for disk vs tail Hα-resolved clumps.However, both the slopes and the P values have to be taken with caution, due to small numbers, especially in the tail clumps.
We find that these distributions are different from those inferred for the Hα clumps of these galaxies detected by (Poggianti et al. 2019) from the MUSE Hα luminosities using the luminosity-size relation by Wisnioski et al. (2012): the expected median size was 440 pc for clumps in the disks and 320 pc for clumps in the tails.Here, the median sizes are ∼ 210 pc, ∼ 211 pc, ∼ 180 pc for disk, extraplanar and tail Hα-resolved clumps, respectively, and ∼ 215 pc, ∼ 223 pc, ∼ 208 pc for disk, extraplanar and tail UV-resolved clumps.
Consistently with what was inferred by Poggianti et al. (2019), clumps in the tails are smaller than those in the disk.Nonetheless, values found in this work are about half the expected size.The origin of this difference is a direct consequence of the differences between luminosity-size relation by Wisnioski et al. (2012) and the one obtained from our HST observations (see Sect. 7).
In the bottom panel of Fig. 14 the slopes of resolved clumps in each category are plotted.Also in the case of SDFs the slope of UV-resolved clumps are smaller (with the exception of disk clumps), even if consistent within errorbars, than those of Hα-resolved clumps.Moreover, disk and extraplanar UV-resolved slopes are almost equal, while in Hα there are hints of a slope increase from disk to tail regions.
The slope increase can be partially explained based on the work in Gusev (2014), whose observations of the nearby galaxy NGC 628 demonstrated that the overall slope of SDFs reaches values between 4.5 and 613 when analysing the smallest structures of the star-forming regions (i.e.what we define as leaves in Sec.4.1) or isolated objects.Instead, the slope decreases up to 2.5 once all the substructures of complex star-forming regions are taken into account.Our trend is analogous.We find steep slopes (∼ 4.4 ± 0.8, consistent with 4.5) in the Hα tails, whose clumps have typically no or few substructures.On the other hand, the slope is smaller in the case of disk clumps, which are more structured than extraplanar and tail clumps.Therefore, including both trunks and leaves in the samples has little effects on the slope of tail clumps, while it may explain the flatter distribution found for disk clumps.Indeed, we observe steeper disk SDFs when using only the leaves (3.28 ± 0.28 in Hα and 3.22 ± 0.23 in UV).Alternatively, recent simulations of star forming regions in presence of different ambient pressures (Nath et al. 2020) found slopes similar to the one of the disk SDFs, while they suggest the presence of a lower pressure environment in the tail.The pressure producing the measured steepening in the tail SDF would be one order of magnitude lower than the typical ICM pressure of our galaxies (Bartolini et al. 2022).Therefore the variation of the slope of the SDF across different environments seems to be different to that expected from environmental effects.
In conclusion, the largest clumps of our sample are found in the disk and in the extraplanar regions of our galaxies, whether we consider UV-or Hα-resolved clumps, and (as hinted by the KS -test), clumps of different spatial categories are likely to follow different SDFs with different slopes.The sizes of the clumps seem to be poorly affected by the environment in which they are embedded, ICM in the tails and ISM in the disks, and more linked to their clustering features.

LUMINOSITY-SIZE RELATIONS
In this section we study the luminosity-size relation, both for Hα-and UV-resolved clumps.Here Hαresolved clumps are BPT-selected to avoid AGN-and LINER-powered regions.To calculate the linear regression fits with the inclusion of an intrinsic scatter, we employed the Python software package linmix (Kelly 2007).linmix implements a Markov Chain Monte Carlo (MCMC) algorithm to converge on the posterior and return a sample of sets of parameters drawn from the posterior distribution.The linear relation fitted by linmix is where L is the luminosity of the clump in the filter in which it is selected, size is the PSF-corrected core diameter, m is the angular coefficient of the correlation, q is the y-axis intercept and G(ε) is the intrinsic scatter, computed from a Gaussian distribution centered in Log(size) + q with standard deviation ε.
In Fig. 17 we plot the datapoints in the (LogL, Log(size)) plane and the best-fitting lines, both for Hα-and UV-resolved clumps (left and right panel, respectively).Clumps are divided according to their spatial position.
Best-fitting parameters are listed in Table 7.The average slope is 2.3±0.4 for Hα-resolved and 1.97±0.17for UV-resolved clumps.The slopes for the disk and extraplanar Hα-resolved clumps are consistent within 1σ and close to 2, while the slope for the tail clumps is steeper.In UV, the slopes of all spatial categories are consistent with each other.
The Strömgren sphere model predicts the slope to be 3 (Beckman et al. 2000), hinting that disk and extraplanar clumps are not well described by this model as a consequence of additional effects to be taken into account, such as RPS, transition from ionization bound to density bound, dust, metallicity and magnetic fields (Wisnioski et al. 2012).On the other hand, Nath et al. (2020) obtained a slope equal to 2 simulating the expansion of ionized bubbles in a Milky Way-like ISM environment (in this case, Hα-resolved tail clumps would be the only ones deviating from the prediction of the model).Cosens et al. (2018) proposed a model explaining why in galactic disks clumps with a low star formation rate surface density (Σ SFR ) seem to follow a steeper relation (slope closer to 3) than clumps with high Σ SFR (slope closer to 2).According to this model, if the expected radius of a Strömgren sphere is larger than the scale-height of the disk (H), the ionized bubble can keep growing only across the galactic plane, according to a power law with slope closer to 2 rather than 3.The flattening occurs only if the ionized region is brighter than a critical value.The fact that our slopes are consistent with 2 might therefore suggest that our clumps have an enhancement in the Hα luminosity, maybe caused by ram-pressure stripping.Such a model would explain also why Hα-resolved tail clumps are likely to follow a steeper relation than clumps in the disk or extraplanar region.Tail clumps are embedded in the spherically symmetric ICM, in place of the gaseous disk of the galaxy, therefore they are not bound by H.

Comparison with previous results
In Fig. 18 we compare the position of our Hα-resolved clumps in the log L − log(size) with those presented in the literature.We show results from Fisher et al. (2017), who study clumps belonging to turbulent, extremely Hα-bright DYNAMO galaxies, and those by Wisnioski et al. (2012), who studied z ∼ 0, isolated, star-forming galaxies (Arsenault & Roy 1988;Kennicutt et al. 2003;Rozas et al. 2006;Gallagher & Hunter 1983;Monreal-Ibero, A. et al. 2007).We also show the best-fitting relations they present in their works.
The luminosities of the DYNAMO clumps in Fig. 18 were corrected re-adding extinction caused by dust, since both our luminosities and those computed by Wisnioski et al. (2012) are not dust-corrected.Dustextincted DYNAMO clumps are then fitted to a power law with slope fixed at 2, as done in Fisher et al. (2017).
As described in detail in Fisher et al. (2017), the radii of DYNAMO clumps were found fitting a 2D Gaussian to the light distribution, with the addition of a constant representing the local background level (Fisher et al. 2017).To make the comparison with DYNAMO as consistent as possible, we derive new PSF-corrected core radii (r gauss ) fitting a 2D Gaussian+constant to our tail Hα-resolved clumps, which are more isolated and in a fainter local background than disk and extraplanar clumps.We then visually select only clumps for which a fit is appropriate.For these clumps, we compute r gauss − r core,corr , finding that it does not correlate with r core,corr , it ranges between 0 and 50 pc and it has a median value of 25.5 pc.Assuming that this difference is a good representation of the value of r gauss − r core,corr for all the Hα-resolved clumps of our sample, we computed a new PSF-corrected core radius r core,corr = r core,corr + 25.5 pc.Therefore the new sizes are 2 r core,corr .The luminosities are re-computed integrating the light within a circle of radius 3 r core,corr .The procedure adopted in Wisnioski et al. (2012) to compute luminosity and size is similar to the one applied in Fisher et al. (2017), though not identical.Therefore we are confident that the corrections we applied to our clumps allow us to make a fair comparison also with the results by Wisnioski et al. (2012).
Our clumps lie between the Fisher and Wisnioski relations, being closer to the one obtained by Fisher et al. (2017), even though they have lower luminosities and sizes compared to the peak of their distribution.With respect to the Wisnioski clump distribution, our resolved clumps are on average larger and, at a given size, brighter.
As shown by Johnson et al. (2017) and Cosens et al. (2018), DYNAMO clumps have both higher SFR and Σ SFR than clumps in isolated spiral galaxies, as a consequence of the starbursty star formation of their hosting galaxies.Being closer to the DYNAMO sample in the luminosity-size relation may suggest our Hα-resolved clumps to have a high Σ SFR , too (hints of that has already been found in Vulcani et al. 2020, in which they studied the resolved SFR-stellar mass relation for the MUSE Hα clumps).

CATALOG
We release the catalogs of Hα-and UV-selected clumps, separately, as online Table .Each clump is univocally determined by the name of the galaxy, a letter (referred to the Astrodendro run in which it has been detected, see Sec.A) and an ID number.We then list the RA and DEC coordinates, the luminosity in the selection filter (not corrected for dust, but corrected for NII in the case of Hα-selected clumps), the morphological quantities (area, major and minor sigma, position angle, core radius and PSF-corrected core radius), the photometric fluxes and their errors in each band, including F680N continuum-subtracted (Hα + NII), a flag for the clump properties in the tree structure (0 trunks which are not leaves, 1 trunks which are also leaves, 2 branches, 4 leaves which are not trunks), a flag for the spatial category (0 tail, 1 extraplanar and 2 disk) and a flag for the BPT classification (0 no BPT diagram available, 1 star-forming, 2 composite, 3 AGN, 4 LINER).Details about how these quantities are computed are given in Secs.4.1 and 4.2.
As an example, in Appendix (Table A3) we report the first ten rows of the Hα-selected clumps catalog.For clarity, for some values not all the significant digits are reported.

SUMMARY
In this paper we have built a sample of star-forming clumps and complexes in six jellyfish galaxies, using a set of HST images in five photometric bands.Clumps were detected independently in UV (∼ 2700 Å) and Hα, in order to probe star formation on different timescales (Kennicutt 1998), while the star-forming complexes were detected from optical emission (∼ 6000 Å) to fully recover the stellar content formed from the stripped material in each clump region.Clumps were also divided in three spatial categories to study separately clumps formed within the disk (disk clumps), clumps likely originated in extraplanar gas but still close to the disk (extraplanar clumps) and clumps formed in the stripped gas out of the galactic disk and embedded in the ICM (tail clumps).Also, clumps in the tail give the unprecedented opportunity to study young stellar populations with no influence or underlying contamination by the stellar disk.
The method we use to detect clumps (Sec.4.1) yields the hierarchical cascade structure of these star-forming regions.While disk clumps are often characterized by .log(L(Hα)) − log(size) comparing our Hα-resolved clumps with those in DYNAMO starburst galaxies (Fisher et al. 2017, blue contours) and those in local, isolated, star-forming galaxies presented in Wisnioski et al. 2012(Arsenault & Roy 1988;Kennicutt et al. 2003;Rozas et al. 2006;Gallagher & Hunter 1983;Monreal-Ibero, A. et al. 2007, black dots).Our clumps are plotted according to their spatial category: disk (red circles), extraplanar (blue stars), tail (green triangles).Our clump luminosities and sizes are corrected in order to make the comparison more trustworthy; for the same reason, DYNAMO clump luminosities are corrected re-adding the effects of dust extinction.The black dashed line is the best-fitting relation by Wisnioski et al. (2012), the blue dashed line is obtained fitting the dust-extincted DYNAMO clumps keeping the slope fixed at 2, as done in Fisher et al. (2017).Clumps in our sample lie in between the two sample, being close in particular to clumps in starburst galaxies.
complex structures where many clumps are localized in bigger structures, extraplanar and tail clumps are typically simple structures with no sub-regions.Moreover, tail clumps tend to be aligned in elongated or arched structures along RPS sub-tails.Interestingly, the Hαselected clumps, UV-selected clumps and F606W starforming complexes are often nested into each other, with Hα-selected clumps being embedded in larger UVselected clumps, and star-forming complexes containing sometimes several UV-and Hα-selected clumps.
We studied the luminosity distribution functions (LDFs), size distribution functions (SDFs) and luminosity-size relations of Hα-and UV-selected clumps as a function of the spatial category.The LDF slopes averaged on all the spatial categories are 1.84±0.03for Hαselected clumps and 1.73 ± 0.09 for UV-selected clumps.The average slopes of SDFs are 3.6 ± 0.6 for Hα-resolved clumps and 3.1 ± 0.3 for UV-resolved clumps.Finally, the average slopes of the luminosity-size relations are 2.3 ± 0.4 for Hα-resolved clumps and 1.97 ± 0.17 for UVresolved clumps.We find no clear difference among disk, extraplanar and tail clumps.The best-fitting slopes of these distributions and relations are consistent among each other, as well as with the results obtained in previous works (Kennicutt et al. 1989;Cook et al. 2016;Santoro et al. 2022) and with theoretical predictions of hierarchical turbulence-driven star formation (Elmegreen & Falgarone 1996;Elmegreen 2006).On the other hand, the luminosity-size relation of the Hα-resolved clumps is more similar to that of clumps in starburst galaxies and therefore it suggests that these clumps, regardless of the spatial category, are experiencing an enhancement in Σ SFR .These preliminary results suggest ram pressure to compress the ISM and increase the Hα luminosity of the clumps, while neither the presence of a disk and its gravity, nor the gaseous conditions of the surrounding medium have a strong impact on the star formation process, once the cold gas cloud conditions are set.
Future works on the mass, age and star formation of the clumps, on trends and gradients with the distance from the galaxies and on their fate will elucidate how and how much these clumps differ from those in undisturbed galaxies, in order to shed light on the effects of ram pressure on the galactic ISM and of environment on star formation.
EG would like to thank A. Ignesti, A. Wolter, A. Marasco, R. Smith, K. George and the GASP team for the useful discussions and comments.We would like to thank also Emily Wisnioski for providing us the data from her paper.This paper is based on observations made with the NASA/ESA Hubble Space Telescope obtained from the Space Telescope Science Institute, which is operated by the Association of Universities for Research in Astronomy, Inc., under NASA contract NAS 5-26555.These observations are under the programme GO-16223.All the HST data used in this paper can be found in MAST: 10.17909/tms2-9250.This paper used also observations collected at the European Organization for Astronomical Research in the Southern Hemisphere associated with the ESO programme 196.B-0578.This research made use of Astropy, a community developed core Python package for Astronomy by the Astropy Collaboration (2018).This project has received funding from the European Research Council (ERC) under the European Union's Horizon 2020 research and innovation programme (grant agreement No. 833824) and "INAF main-streams" funding programme (PI B. Vulcani).

APPENDIX
A. ASTRODENDRO PARAMETERS SETTING Three parameters regulate how Astrodendro builds the tree structure: • min value: the algorithm stops when the flux threshold reaches this value, instead of zero; • min npix: minimum number of pixels for a clump to be included to the tree structure; • min delta: the threshold is not lowered in a continuum way, but at steps of min delta.If no min delta is given, the algorithm identifies each local maximum as a new sub-clump.min delta should be high enough to avoid the detection of noise peaks in the surface brightness distribution as sub-clumps.
We performed three runs of Astrodendro for the F275W and Hα images of each galaxy, adopting the following parameters (min npix= 5 in all the runs): • RUN A: min value= 2.5σ; min delta= 5σ; • RUN B: min value= 2.5σ; no min delta.Given that a clump candidate is detected only if its brightest pixel is brighter than about min value+min delta, regions for which each pixel has counts between min value and min value+min delta are not detected.Since we want to detect also these fainter clumps, we ran Astrodendro a second time without defining min delta.This run is executed on an image masked for the clumps detected in run A and only trunk clumps are retained, to avoid including spurious local maxima.
• RUN C: min value= 2σ; no min delta.This run is performed on an image masked for the clumps detected and kept in runs A and B. For the same reasons explained for run B, we kept only the trunks clumps of run C. Also, as a consequence of removing the high-frequency components of the image, denoising introduces a sort of smoothing, and part of the light of the brightest regions of the image, already detected as clumps, may eventually smooth out of the masks defined from runs A and B. Thus, even masking the image for the clumps Table A3 (continued).First ten rows of the Hα-selected clumps catalog.Columns from 18 to 23: uncertainty on the density flux in the filter F336W (errF336W), density flux and uncertainty in the filter F606W (F606W and errF606W), density flux and uncertainty in the filter F680N (F680N and errF680N), density flux in the filter F814W (F814W).

Figure 3 .
Figure 3.Comparison between the isophotal radius (riso) and twice the PSF-corrected core radius (2rcore,corr), defined in Sec.4.1.1,both for Hα-resolved clumps (left panel) and UV-resolved clumps (right panel).The grey dashed line is the 1 : 1 relation, while the red solid line is the best-fitting line.The best-fitting line is in good agreement with the 1 : 1 relation, both for Hα-and UV-resolved clumps.

Figure 4 .
Figure 4. Flow chart summarizing the selection procedure adopted in this paper to confirm (green checkmark) or reject (red cross) clump candidates detected by Astrodendro.

Figure 5 .
Figure 5. Images of 4 Hα clumps of JO206 in all the filters.Each clump is shown in 6 different filters, which are (from upper left to lower right): F275W, F336W, F606W, F680N, F814W, Hα.Each filter is labelled in green or red whether we have a detection or not, according to our definition (Sec.4.2).The FOV is constant for all the clumps (a lengthscale equal to 250 pc is plotted in lower left panel of the third clump).

Figure 6 .
Figure 6.Map of the clumps detected in JO201, superimposed onto the image in the filter used for the detection.Upper panels: Hα-selected clumps.Lower panels: UV-selected clumps.Left panels: field of view including all the clumps.Right panel: zoomed-in version on the vicinity of the disk (highlighted in the left panel with the black dashed rectangle).Colors in the right panels represent the spatial category and the tree structure (Sec.3 and 4.1): disk clumps are plotted in red (trunks which are not leaves), orange (trunks which are leaves) and yellow (leaves which are not trunks).Similarly, extraplanar clumps are plotted in dark blue, blue and light blue, and tail clumps are plotted in dark green, green and light green in right panels.The grey dashed contour is the galaxy disk contour (see Sect. 3. In the left panels, for the sake of clarity, the tail clumps are plotted as green dots of fixed size.The regions highlighted and labelled as A, B, C and D are shown in figure 7.

Figure 7 .
Figure 7. Hα images of the JO201 regions highlighted in the upper right panel of figure 6.The colors are the same of figure 6.

Figure 8 .
Figure 8. Zoomed-in F606W image of some star-forming complexes and clumps in JO201.Hα-selected clumps are plotted in dark orange, UV-selected clumps in violet and complexes in dark violet.

Figure 9 .
Figure 9. Histograms of the number of star-forming clumps and complexes in each galaxy, divided according to the selection filter and the spatial category.For each galaxy three panels are shown, with the number of Hα-selected clumps (left panel), UV-selected clumps (middle panel) and star-forming complexes (right panel), divided according to their spatial category: disk (red), extraplanar (blue) and tail (green).

Figure 10 .
Figure10.Fraction of Hα-selected (top row) and UV-selected (bottom row) clumps per spatial category and per galaxy.In each row, from left to right: disk, extraplanar and tail.Y-axis is normalized for the number of clumps in the galaxy and in the spatial category.Notice that the Hα luminosity of the Hα-selected clumps is the integrated emission of the Hα line, therefore in erg/s, while the UV luminosity of the UV-selected clumps is in erg/s/ Å.
of the best-fitting values of the LDFs and SDFs when fitted to the different samples of star-forming clumps and complexes.Column (1) refers to the clump selection photometric band.Column (2), from top to bottom: disk (D), extraplanar (E) and tail (T) sub-samples of Hα-selected and UV-selected clumps; the last row refers to starforming complexes, which are only in the tails by construction.Columns (3), (

Figure 12 .
Figure 12.Luminosity distribution functions d N/dL of Hα-selected (upper panels) and UV-selected (lower panels) clumps.Clumps are divided according to their spatial category: disk (left panels, in red), extraplanar (middle panels, in blue) and tail (right panels, in green).For each plot we show: the empirical LDF of the corresponding sample (open circles with errorbars), generated with equal-number bins (i.e. each bin contains the same number of objects, seeCook et al. 2016), and the best-fitting line (dashed line).Notice that the Hα luminosity of the Hα-selected clumps is the integrated emission of the Hα line, therefore in erg/s, while the UV luminosity of the UV-selected clumps is in erg/s/ Å.

Figure 14 .
Figure 14.Comparison of the slopes of the LDFs (left panel) and SDFs (right panel) of star-forming clumps and star-forming complexes, both as a function of the selection band and the spatial category (see Table5).Circles are Hαselected clumps, squares are UV-selected clumps and triangles are star-forming complexes.Colors refers to the spatial category: red for disk, blue for extraplanar and green for tail.
obtained when a set of power laws are fitted to the disk LDFs divided in intervals (Sect.6.1).From left to right: clump selection photometric band (Phot.band); best-fitting slope α and luminosity range boundaries of the interval Lmin and Lmax (Parameters); names of the intervals (Faint-end, Plateau, Bright-end and Secondary-peak).

Figure 15 .
Figure 15.Luminosity distribution functions d N/dL of Hα-selected (upper panels) and UV-selected (lower panels) clumps.Clumps are divided according to their spatial category: disk (left panels, in red), extraplanar (middle panels, in blue) and tail (right panels, in green).For each plot we show: the empirical LDF of the corresponding sample (open circles with errorbars), generated with equal-number bins (i.e. each bin contains the same number of objects, seeCook et al. 2016), and the best-fitting line (dashed line).Notice that the Hα luminosity of the Hα-selected clumps is the integrated emission of the Hα line, therefore in erg/s, while the UV luminosity of the UV-selected clumps is in erg/s/ Å.

Figure 16 .
Figure 16.Size distribution functions for disk (red), extraplanar (blue) and tail (green) clumps.Top: Hα.Bottom: UV.Resolved clumps are shown as empty circles with 1σ errorbars, while unresolved clumps are plotted as filled circles without errorbars.In this case, SDFs are not normalized by the total number of clumps and the x-axis is in linear scale.

Figure 17 .
Figure17.Luminosity-size relations for Hα-resolved clumps (on the left) and UV-resolved clumps (on the right).The clumps are plotted according to their spatial category: disk (red circles), extraplanar (blue stars), tail (green triangles).The best-fitting lines to the three categories are plotted as solid lines of the corresponding color.The shaded areas are the uncertainties on the fits at 2σ.Note that Hα luminosity is in erg/s, while F275W in erg/s/ Å.

Figure 18
Figure 18.log(L(Hα)) − log(size) comparing our Hα-resolved clumps with those in DYNAMO starburst galaxies(Fisher et al. 2017, blue contours) and those in local, isolated, star-forming galaxies presented inWisnioski et al. 2012(Arsenault & Roy 1988;Kennicutt et al. 2003;Rozas et al. 2006;Gallagher & Hunter 1983;Monreal-Ibero, A. et al. 2007, black dots).Our clumps are plotted according to their spatial category: disk (red circles), extraplanar (blue stars), tail (green triangles).Our clump luminosities and sizes are corrected in order to make the comparison more trustworthy; for the same reason, DYNAMO clump luminosities are corrected re-adding the effects of dust extinction.The black dashed line is the best-fitting relation byWisnioski et al. (2012), the blue dashed line is obtained fitting the dust-extincted DYNAMO clumps keeping the slope fixed at 2, as done inFisher et al. (2017).Clumps in our sample lie in between the two sample, being close in particular to clumps in starburst galaxies.

Table 1 .
Summary of the main properties of the galaxies studied in this paper and of their host clusters.

NO YES NO Rejected Confirmed NO -3611 Hα -2293 UV -4 Hα -30 UV -69 Hα -186 UV 6090 Hα 6259 UV
. If a match is found, the HST clump is validated; if not, the clump is validated only if either the redshift from

Table 3 .
Number of clumps in each sub-sample.In brackets, the number of clumps in the each sample, but selected in order to avoid regions powered by AGN emission (see Sect. 4.2).
Note-Number of clumps detected in each galaxy and depending on the spatial category.From left to right: photometric band in which the clumps were detected (column 1), name of the galaxy (2), number of LT clumps (3), number of disk-LT clumps (4), number of extraplanar-LT clumps (5), number of tail-LT clumps (6), number of resolved clumps (7), number of resolved-disk clumps (8), number of resolved-extraplanar clumps (9), number of resolved-tail clumps (10).

Table 4 .
Number of star-forming complexes detected in the tails of each galaxy.

Table 5 .
Luminosity and size distribution functions best-fitting parameters.

Table 6 .
Best-fitting slopes to the intervals of the disk LDFs.