The Local Ultraviolet to Infrared Treasury I. Survey Overview of the Broadband Imaging

The 22 LUVIT NUV (GO-15275) galaxies were chosen for their high legacy value. They span a significant range of metallicity, gas fraction, SFR intensity, and environment. Several criteria were implemented in order to restrict the sample from the $\gtrsim 100$ galaxies in the Local Volume. Building off the foundational work of ANGST, as well as earlier HST optical campaigns (Tully et al., 2006; Dalcanton et al., 2009), the LUVIT sample was restricted to galaxies with distances less than 3.5 Mpc¹¹1Since the selection of the LUVIT sample, revised distance estimates for one of the targets, UGCA292, place it at a distance greater than the 3.5 Mpc cutoff., to mitigate the effects of stellar crowding. Furthermore, only galaxies with estimated stellar masses $\lesssim 10^{7.5}$ $M_{\odot}$ were included in order to target the lowest mass, lowest metallicity star-forming systems in the local universe. The sample was further restricted to galaxies with existing optical HST data in at least two broadband filters, further requiring the data be of sufficient depth to (i) constrain main sequence (MS) stars with masses $>5$ $M_{\odot}$ ($M_{V}\sim-0.5$), which are responsible for $\gtrsim 70\%$ and $\gtrsim 60\%$ of the FUV and NUV flux, respectively (Johnson et al., 2013), and (ii) detect and measure red giant branch (RGB) stars $>2$ magnitudes fainter than the tip of the RGB. These fainter RGB stars are needed to anchor older SFR estimates and to provide a dense network of dust extinction probes. Finally, a galaxy was also required to have existing, rich ancillary datasets. Each galaxy was required to have H$\alpha$ and Hi, many were observed by GALEX and Spitzer, and most have also been targeted by other major facilities across the electromagetic spectrum, e.g., Herschel, VLA, CARMA, and IRAM.

Following the NUV observations, we obtained additional NIR and select optical imaging in GO-16162 for all but three of the most distant LUVIT NUV targets (DDO6, M81-DWARF-A, and UGC8760). The omission of these three galaxies substantially reduced the orbital request of the program with minimal impact to the science goals of the GO-16162 observations.

Implementing the above four criteria resulted in a sample that spans the remaining parameter space in the specific SFR – stellar mass plane which was not covered by LEGUS or existing Local Group surveys. Figure 1 compares the specific star formation rate (SSFR) as a function of stellar mass ($M_{*}$) of the LUVIT and LEGUS samples against a broader sample of star-forming galaxies within the Local Volume. The Local Volume sample is that of Lee et al. (2009), and includes all star-forming galaxies within 11 Mpc of the Milky Way. The SFR and $M_{*}$ values for the LEGUS galaxy sample were taken from Calzetti et al. (2015, Table 1) for the galaxies observed by LEGUS, and taken or derived from Lee et al. (2009, Table 1) for the galaxies for which LEGUS used archival datasets. In both the 11 Mpc Local Volume sample and the LEGUS sample, the SFR was calculated from the GALEX FUV flux after correcting for dust attenuation. Following Lee et al. and Calzetti et al. (2015), the stellar mass estimates are calculated from the absolute $B$-band magnitude of each galaxy, using a morphology-color relationship determined by Bothwell et al. (2009) and mass-to-light ratio models by Bell & de Jong (2001). The values for LUVIT were similarly derived from Lee et al. where possible, or from the Updated Nearby Galaxy Catalog (UNGC; Karachentsev et al., 2013; Karachentsev & Kaisina, 2013). The stellar masses for the Local Group galaxies (M33, M31, LMC, and SMC) are taken from Sick et al. (2015, M31) and McConnachie (2012), while the SFRs are taken from the UNGC (Karachentsev & Kaisina, 2013).²²2For three galaxies (UGC8508, LMC, and SMC), a FUV SFR was not available in any of the above references and the H$\alpha$-based SFR from Karachentsev & Kaisina (2013) is plotted instead. The LUVIT and LEGUS galaxies are color coded by distance, with the color map diverging at 3.5 Mpc, the approximate upper and lower limit of the LUVIT and LEGUS surveys, respectively. As described above, the LUVIT galaxies were chosen as representatives of the lowest stellar mass, star-forming galaxies in the local universe.

Table 1 lists basic properties of the LUVIT galaxies, as well as commonly used alternate names, the filters obtained as part of LUVIT (from both GO-15275 and GO-16162 programs), and the archival filters included in the LUVIT reductions. As noted, many of the galaxy measurements are taken from the Catalog and Atlas of the LV galaxies (LVG; http://www.sao.ru/lv/lvgdb/), which is the online portal of the Updated Nearby Galaxy Catalog (Karachentsev et al., 2013). Interested readers are referred to either resource for detailed descriptions of the measurements or derivations of each quantity.

Upon execution of the UV observations for each target, we searched the MAST archive to ensure we identified all relevant archival observations. In particular, we identified, for each GO-15275 pointing, the maximally overlapping archival imaging in broad optical or NIR bands, from any of Hubble’s primary imaging cameras: WFPC2, ACS/WFC, WFC3/UVIS or WFC3/IR. These datasets were subsequently included in our simultaneous, multiband reductions (Section 3). While data of sufficient depth in at least two optical broadband filters was a requirement for inclusion in the sample (Section 2.1), a quarter of the sample galaxies also have archival WFC3 NIR imaging (the majority from ANGST-SNAP; Dalcanton et al., 2012b).

In general, we did not include shallower, partially overlapping archival data when deeper, fully overlapping data existed in the same (or similar) bands. In some targets, deeper data with newer cameras (e.g., ACS/WFC or WFC3/UVIS) substantially overlapped significantly shallower data obtained with WFPC2: in these cases, we used only the ACS or WFC3/UVIS data⁵⁵5In addition to deeper photometry, the ACS and WFC3/UVIS instruments have fewer chips and associated chip gaps, and simpler footprints, than WFPC2, reducing the spatial areas with differing and complex completeness functions in the multiband stellar catalogs.. However, for four targets, archival observations from different programs taken in the same filter and camera combinations were included. The data from the multiple programs either increased the spatial area of the archival imaging in a given filter relative to the UV observations (Sextans A, UGC7577) or substantially increased the exposure time of the archival imaging (NGC4163, SAGDIG).

In addition, four targets had deep medium-band imaging in the archive (IC1613, SAGDIG, Sextans A, and Sextans B), which we also included in our multiband reductions. The medium-band archival imaging was in F127M, F139M, and F153M with WFC3/IR for SAGDIG, Sextans A, and Sextans B (GO-14073, PI M. Boyer) and in F390M with WFC3/UVIS for IC1613 (GO-12304, PI J. Holtzman).

For a small number of targets, existing archival optical imaging was extremely deep (DDO210, LEO-A, SAGDIG, and WLM-POS2). Including all the available archival exposures would have resulted in a depth well beyond that which is necessary for the goals of the LUVIT program, and would have needlessly increased the required computational resources for analyzing these targets. Therefore, in these cases we limited the archival imaging to the first full dither pattern⁶⁶6For Sag-DIG, the deep imaging (GO-12273; PI R. van der Marel) was in F814W only and taken at 3 primary positions, each with its own small dither pattern. Since the deep observations overlapped previously existing F814W observations (GO-9820; PI Y. Momany) we included the full small dither pattern of the first of the primary positions, choosing better PSF sampling over increased chip gap coverage. of the deep archival observations in each band, ensuring we recovered the best possible sampling of the point s pread function and coverage of chip gaps. This strategy still resulted in depths in the archival bands which exceed that of the typical target in the LUVIT survey.

We included data obtained concurrent to and after the GO-15275 observations in one target. Partially overlapping WFC3/UVIS observations in Sextans A were obtained by GO-16104 (PI J. Roman-Duval), in the F225W, F275W, F336W, F475W, and F814W filters. These observations were designed to support the Hubble Director’s Discretionary program ULYSSES, which obtained UV spectra of massive stars in the local universe (Roman-Duval et al., 2020). The GO-16104 observations extend the UV and optical coverage of the star-forming region of Sextans A, with the footprints of the observations including the marginal IRAM 30 m telescope CO detection by Shi et al. (2015). All the GO-16104 observations are at the same pointing; their position in Figure 2 is represented by the archival UV observation footprint located below and partially overlapping with the LUVIT observation footprints.

The total exposure time in the archival bands varies widely, as did the science goals of the various GO programs. Table 4 provides details of the archival observations used in the LUVIT reductions.

Both GO-15275 and GO-16162 obtained parallel imaging along with the primary observations. While we do not present the results of the parallel imaging in this survey overview, for completeness we summarize the observational strategies used in obtaining the parallel exposures below.

Optical ACS/WFC images in F606W and F814W were obtained in parallel for all of the GO-15275 pointings. The parallels were obtained to enable the spatial extent to be determined and the ages of any extended stellar populations in the LUVIT galaxies to be estimated. In cases where there is no discernible stellar population associated with the target galaxy, the parallel imaging can be used to provide an estimate of the foreground and background populations present in the primary pointings. In the GO-16162 program, parallels in F606W and F814W were obtained for the 6 galaxies (7 targets) where the ACS field fell within $3r_{25}$ (DDO210, IC1613, Sextans A, Sextans B, UGC6817, and both pointings of WLM). The location, orientation, and dither pattern of the parallel observations were set by the location, orientation, and dithers of the primary pointings.

In GO-15275, the F606W and F814W filters were each observed for one full orbit for every target, obtaining a full dither pattern in each filter. For three-orbit targets, the third orbit was split between F606W and F814W, with an exposure in each filter obtained at each dither position. As with the primary observations, the exposure times were determined using the orbit planner in APT, maximizing the total exposure time within the orbital visibility window. The resulting exposure times produce deep optical CMDs.

A similar observational strategy was followed for the parallels obtained in GO-16162. ACS/WFC parallels were obtained in conjunction with the WFC3/IR observations. For the targets with one WFC3/IR orbit (IC1613, Sextans B), the parallels were split between the F606W and F814W filters, with longer exposures on F814W. For the targets with at least two WFC3/IR orbits (DDO210, UGC6817, Sextans A, both WLM positions), F606W and F814W were observed in parallel for one orbit each. UGC6817 had three WFC3/IR orbits; in this case, as in GO-15275, the third orbit was split evenly between F606W and F814W.

The data reduction pipeline starts from flc files for WFC3/UVIS and WFC/ACS observations, flt files for WFC3/NIR observations, and c0m/c1m files for WFPC2 observations, all downloaded from MAST. First, the images in each band from each individual visit are run through astrodrizzle to update the data quality extensions to improve cosmic ray masking. Then, each individual image is masked and calibrated for distortion using the DOLPHOT preparation tasks. After these preparation steps, PSF photometry is measured separately on each individual detector from each individual exposure. The resulting catalogs for all of these individual detector runs are then cross-matched across all of the exposures to refine the image world coordinate system of all of the headers (Williams et al., 2014), aligning all of the images covering the region to a precision of $\sim$0.1 pixel. Once the headers of the original flc, flt, and c0m images have been updated with the aligned world coordinate systems, the preparation process, including astrodrizzle and the DOLPHOT preparation tasks, are rerun separately for each band but including all visits in the astrodrizzle stack to optimize cosmic ray masking.

For each target, a drizzled reference image was created from all exposures in the chosen reference camera and filter combination. For the majority of the LUVIT reductions (all the targets included in GO-16162), the drizzled image created from the exposures obtained in F475W (from either WFC3/UVIS or ACS/WFC, depending on the target) was specified to be used as the reference image. The F475W band is the most likely to contain a large number of both red and blue stars, allowing it to be used to successfully align NIR images as well as NUV images. For all GO-16162 targets, the F475W imaging was sufficiently deep to provide good alignment.

Three targets in GO-15275 were not included in GO-16162 and used a different filter for the reference image. The camera and filter used for creation of the reference image for these targets were WFC3/UVIS F275W (DDO6, UGC8760) and ACS/WFC F555W (M81-DWARF-A). If not specified (as was done for the targets included in GO-16162), the drizzled image produced from the filter with the greatest total exposure time is chosen by the pipeline as the default reference image. For DDO6 and UGC8760, this was the drizzled F275W image. Without the need to align both NIR and NUV imaging, this resulted in successful multiband alignment for DDO6 and UGC8760. For M81-DWARF-A, the most successful alignment was found using the drizzled F555W image as the reference image, rather than the pipeline’s default choice of F814W.

Finally, we note that, as shown in Figure 2, the star-forming region of many of the LUVIT targets fits primarily within one WFC3/UVIS chip. In one case (DDO6), a lack of NUV sources on the one WFC3/UVIS chip that does not include the star-forming region resulted in a failure to find any acceptable astrometric solution between the UV and optical bands for that chip. As a result, the reduction pipeline was run on only one of the two WFC3/UVIS and ACS/WFC chips (chip 2 in both cases). Therefore, the photometric catalogs for DDO6 include only data from the WFC3/UVIS and ACS/WFC chips containing the star-forming region. In general, the relative alignment between individual exposures and the drizzled reference image will have larger relative alignment uncertainties in regions of very low source density, where few stars may be matched between the individual image and the reference image.

After the initial image processing and alignment described above, all images covering the region are put into DOLPHOT together to obtain, over the area covered by the bounds of the reference image (Section 3.1), full-depth photometry catalogs that include measurements of all stars in all images, as well as the resulting combined measurement for every measurement in each band. When DOLPHOT runs on a full list of aligned images with multiple bands including multiple cameras, it generates a stack of all of the image pixels in memory and searches the stack for peaks at least 3-$\sigma$ above the noise in the full stack including all bands. It then forces a fit of the point spread function at the locations of all these peaks in every individual image, accounting for all neighbors. It does this iteratively by fitting the stars, subtracting them from the images, and searching the residuals for stars missed in the previous pass. It also compares the flux measured with PSF and aperture photometry for the most isolated stars in the data to correct the photometry for PSF-related systematics. The result is a measurement of the flux (or upper limit) of every source in the catalog from every image in the stack, as well as a combined measurement of the flux in each filter, taking advantage of multiple exposures in each band to refine the flux. Table 5 lists the values of the user-specified DOLPHOT parameters utilized in the LUVIT reductions.

The pipeline produces the standard, comprehensive, DOLPHOT output catalogs (‘.phot’ files; described in detail in the DOLPHOT documentation⁷⁷7http://americano.dolphinsim.com/dolphot/). These catalogs include, for each source detected, the measured fluxes from the PSF fit for that source’s location in each image, the calculated Vega system magnitude, the associated photon-count-based uncertainties, the signal-to-noise ratio, and metrics which indicate the quality of the PSF fit for that source (including sharpness, roundness, crowding, and $\chi$). Combined measurements for each source, in each filter, are also included in the DOLPHOT output catalogs.

We subsequently cull these photometric catalogs using the GST quality criteria established by Williams et al. (2014), and summarized in Table 6. First, sources must be detected with a signal-to-noise ratio $>4$. Second, we apply camera-specific quality criteria to the combined measurements in each filter, using the metrics related to the quality of the PSF fit. The limiting values for these measurements were chosen to balance the goals of retaining a very high fraction of total measurements while removing a high fraction of the sources that lie outside the stellar features in the CMDs (Williams et al., 2014).

Specifically, we remove likely cosmic rays and background galaxies using the sharpness parameter, which provides a measure of the relative concentration of flux in the central vs. outer pixels compared to the PSF. High sharpness values (high central concentration) are typically caused by cosmic rays or hot pixels, while low values typically indicate background galaxies or blended stars. The crowding parameter provides a measurement of the extent to which a source’s PSF fit was affected by neighboring sources. Larger crowding values indicate a higher density of other sources in the region of the measured source’s PSF profile, and a higher likelihood that the photometric measurements have higher systematic uncertainties caused by imperfect subtraction of neighboring sources’ PSF fits. Finally, $\chi$ provides a measurement of the goodness-of-fit of the PSF to the source’s flux distribution.

In addition to these GST quality criteria, we also remove detections that are associated with diffraction spikes, which are artifacts resulting from saturation of foreground bright stars. We do this by identifying areas with a local stellar surface density that exceeds the median stellar surface density of the immediately surrounding regions by $>$3–5$\sigma$, where $\sigma$ is defined as the expected Poisson noise (the square root of the local stellar surface density). The 3–5$\sigma$ threshold prevents us from removing bona-fide stars in crowded star-forming regions, while effectively eliminating objects associated with diffraction spike features. The threshold was set on a target-by-target basis after visually inspecting the resulting sources identified for removal as diffraction spike artifacts. We use a 3$\sigma$ threshold for NGC 4163, SagDIG, and WLM Position 1 and a 5 $\sigma$ threshold for Leo A, Sextans A, Sextans B, and WLM-POS2; we use a 4$\sigma$ threshold for the remainder of the targets.

The above cuts are demonstrated in Figure 3, which shows the spatial position of sources in DDO210 that pass and fail the GST criteria, as well as the additional diffraction spike artifact criterion, and compares the optical CMDs of sources that pass all quality criteria to those that are removed as likely diffraction spike artifacts.

Given the large range of wavelengths covered by the NUV to NIR observations, we do not require a source’s measurements to pass the criteria simultaneously in all filters. Rather, we assess these criteria using the combined measurements for the filter(s) being analyzed. Unless noted otherwise, all data presented in this paper pass both the GST quality and diffraction spike artifact criteria in the relevant filter(s).

Figure 4 shows the distribution on the sky of sources passing all stellar quality criteria (Section 3.3) in NUV, optical, or NIR (when available) filters, and in various combinations thereof. The maximum spatial extent of the stellar catalogs is set by the extent of the drizzled reference image used for each target (Section 3.1), and is largest when the drizzled reference image is based on ACS/WFC observations.

The diversity in spatial scale of the LUVIT galaxies, relative to the HST instruments’ fields of view, can be seen in the spatial distributions of the observed stars. More distant and/or compact galaxies, for which the LUVIT observations cover the entirety, or the vast majority, of the galaxy, show a clear drop in stellar density over the field of view of the LUVIT observations. In other targets, typically those within the Local Group, the LUVIT pointings cover only a fraction of the galaxy and the stellar density remains relatively constant throughout.

In the galaxies where the field of view of the HST observations is large enough to contain the bulk of the stellar population of the galaxy, the relatively compact spatial distribution of stars passing our quality criteria in both NUV filters reflects the fact that the majority of the most recent SF in each galaxy is found in the innermost regions. Older RGB stars, detected in the optical and NIR, are more numerous and have a larger spatial extent than the brightest, young stars observed in the NUV. The optical images are typically the deepest images, reaching further down the luminosity function than other observations, and providing greater sensitivity to low surface brightness populations. While old red giant branch stars are the dominant population of stars detected in the NIR bands, the NIR images are typically shallower than the optical images, with a smaller field of view (Figure 2).

In this section, we present CMDs highlighting different aspects of the panchromatic LUVIT stellar catalogs. Only sources meeting all quality criteria (Section 3.3) are shown in the CMDs.

Figure 5 shows CMDs in the NUV, optical, and NIR for each target, derived from the full stellar catalogs. For targets with more than two optical (or NUV) broadband filters, multiple optical (NUV) CMDs are shown. To demonstrate the increased separation between stellar populations with increased color baseline, when multiple CMDs in the optical (or NUV) are shown, each uses the longest wavelength filter when computing the color and for the magnitude. A sequence of massive, young stars is seen in each of the NUV and optical CMDs. Many LUVIT galaxies also display strong He-burning sequences (notably IC1613, Leo A, NGC3741, NGC4163, UGC7577, UGC8508, UGC8651, UGC9240, and WLM), as well as significant AGB populations. All display strong features associated with older stellar ages, such as the red giant branch, and for deep enough optical observations, the horizontal branch and red clump.

The optical images typically reach the faintest apparent magnitudes, while the NUV and NIR images are shallower. Due to the heterogeneous nature of the archival data, a few targets with particularly complex archival datasets (Section 6) have limiting depths in the optical filters that vary significantly depending on the spatial position of a star. For targets with particularly large differences in exposure times in a given filter as a function of sky position, this results in clear artifacts in the CMDs. The most egregious example is in Sextans A, in the F555W and F814W (WFPC2) filters.

Most LUVIT targets are well removed from the MW disk plane (with absolute Galactic latitudes $b\gtrsim 40$^∘; Table 1), and are thus expected to have relatively low MW foreground contamination in their CMDs. The lowest Galactic latitude LUVIT galaxies are SagDIG ($b=-16.2879$), Antlia ($b=22.3124$), DDO 210 ($b=-31.341$), and M81 Dwarf A ($b=33.0119$). In particular, the CMDs of SagDIG show significant foreground contamination; in addition to having a low Galactic latitude, SagDIG is also in the direction of the Galactic center.

Figure 6 presents, for each target, individual stars in each CMD color coded by the total number of filters in which the star passes all quality criteria. To provide additional information content in this figure, for targets with more than two optical (or NUV) filters we plot one CMD for each of the adjacent optical (NUV) color combinations. Only stars in spatial regions of the sky with full-filter coverage (Section 5.1; Figure 6, left-most panel) are included in these CMDs. For most targets, this is limited by the field of view of the WFC3/IR observations, and is a significant fraction of the footprints of the UV and optical observations (Figure 2). However, for targets SEXTANS-A and WLM-POS-1, the heterogeneity of the archival datasets result in an unusual, and highly restricted, spatial area with full-filter coverage. For WLM-POS-1 the effect is so severe that instead of showing the true full-filter region, a “full wavelength coverage” region is instead shown. This region is computed by requiring observations to be possible in all filters with the exception of F439W, which is the filter which most severely restricts the spatial area (Section 6).

Nearly all stars passing the quality criteria in both UV filters are also detected in at least one other broadband filter. In fact, most stars detected in both UV filters have measurements in the majority of the available filters. The stars with measurements in the greatest number of filters preferentially populate the upper MS and BHeB sequences. Brighter RGB stars and AGB stars are reliably observed in all optical and NIR filters. A significant population of stars at the faint end of the CMDs are detected in only the optical or the NIR bands.

Figure 7 shows, for two LUVIT galaxies, the location in UV, optical, and NIR CMDs of individual stars in four stages of stellar evolution (upper end of the MS, BHeB, RHeB, and RGB), as well as the flux rates measured in each observed filter for each of the stars. Note that due to the nature of the forced photometry (Section 3.2), a flux rate is measured at the location of an observed source in all images (and can be negative). However, the rightmost panel of Figure 7 only displays a flux rate measurement when it translates into a detection passing all quality criteria. Notably, for the RGB stars that do not appear in the NUV CMDs due to negative flux measurements in F275W, we do not plot their flux rate measurements in F275W in the right-most panels.

Figure 8 demonstrates more clearly which stellar populations are typically observed in the optical and NUV, in the optical and NIR, or in all three wavelength regimes. Figure 8 shows the optical CMD distribution of stars, within the full-filter region, whose measurements pass all quality criteria in four or more filters ($N_{\rm filter}\geq 4$) for each of the targets with broadband NIR imaging. To be included in Figure 8, the star must be measured in the two deepest optical filters for that target, with the additional measurements required to be in either the LUVIT NUV and/or the NIR filters. Thus, for a star shown in Figure 8 as observed in four filters, the two nonoptical measurements may be in the NUV or in the NIR, or one may be in each.

The stars observed in at least six filters (required for Figure 8 to be the two deepest optical, two NUV, and two NIR broadband filters) are typically the brightest and most massive upper MS and BHeB stars. The stars with measurements in five filters (required for Figure 8 to be those that are observed in the two deepest optical and any three of the four LUVIT NUV and NIR broadband filters) are primarily the fainter (less massive) MS and BHeB stars, as well as RHeB stars. The majority of stars with measurements in four filters are observed in the optical and in either the NUV or the NIR (rather than in one of each of the NUV and NIR bands). Contours in the right-most panel of Figure 8 differentiate the CMD distributions of stars detected in two NUV and two optical bands (solid cyan contours; primarily MS and BHeb stars) and in two NIR and two optical bands (dashed coral contours; primarily RHeB, RGB, and AGB stars).

For the three targets without archival or GO-16162 NIR observations (DDO6, M81 Dwarf A, and UGC8760), ASTs were run only in the full-filter region, which in these three cases provides good coverage of the optical extent of the galaxy. For these targets, the full-filter region was determined using a convex hull algorithm applied to the set of stellar measurements for which the photon count rate was nonzero in all filters. The input AST lists were created only within the spatial boundary identified as the full-filter region.

For the remainder of the LUVIT targets, ASTs were run over the full area of the reference image. In this case, the full-filter region was determined using a convex hull algorithm applied to the set of ASTs for which the count rate was non-zero in all filters. For a subset of these targets (Leo A, Sextans A, Sextans B, UGC 6817, UGC 7577, UGC 8651, UGC 8833, WLM Position 1), the full-filter region was sufficiently complex that the convex hull algorithm was insufficient. In these cases, the spatial boundaries resulting from the convex hull algorithm were modified by hand to accurately capture features such as the corners of the WFPC2 “sawtooth” footprint (caused by the boundary between the Wide-Field Camera and Planetary Camera) and/or chip gaps that were not covered with dithered observations in one or more filters. Once the boundary of the full-filter region was established, all ASTs contained within the spatial boundary of the full-filter region were used for the AST-based analyses included below.

For each target, input ASTs are generated simultaneously in all available bands using the capabilities provided by the Bayesian Extinction and Stellar Tool (BEAST; Gordon et al., 2016), which is an open-source project.⁸⁸8https://github.com/BEAST-Fitting The BEAST maps a stellar evolution library onto a stellar atmospheric model based on an input stellar initial mass, age, metallicity, effective temperature, and surface gravity. To generate the list of input ASTs, we populate a physics model grid by varying stellar and dust parameters. We use PARSEC version 1.2S isochrones (Bressan et al., 2012; Tang et al., 2014; Chen et al., 2015) for the stellar evolutionary library. For the stellar atmospheric models, we adopt the grid by Castelli & Kurucz (2003) for local thermodynamic equilibrium (LTE) models (lower temperature stars; spectral type A and cooler) and the TLUSTY OSTAR and BSTAR grids by Lanz & Hubeny (2003, 2007) for non-LTE models (higher temperature stars). The BEAST provides implementation of a smoothly varying range of possible dust properties (from SMC-like to MW-like dust curves) via the amount of dust extinction ($A_{V}$), the average dust grain size ($R_{V}$), and the strength of the 2175Å  bump. Combinations of requested dust properties are applied to the intrinsic stellar spectral energy distributions (SEDs) computed from the stellar evolutionary library and stellar atmospheric models to compute extinguished SEDs.

For each galaxy, we first create a stellar physics model grid, with dust effects applied, that appropriately covers the galaxy’s entire potential stellar and dust parameter space. We divide the range of fluxes in the physics model grid into 40–70 linear bins in magnitude space, depending on each galaxy’s physics model coverage, leading to a typical flux bin size of 0.3 – 0.6 magnitude in each filter.

To capture nonlinear changes in photometry characteristics with background source density, we also create stellar number density maps with 5″$\times$5″ resolution. We use a catalog of all sources that pass the GST criteria in two select optical filters,⁹⁹9For the primary LUVIT reductions including GO-15275, GO-16162, and archival data, these filters are F475W and F814W. Of the three targets not included in GO-16162, the optical filters were F475W and F814W for DDO6, F555W and F814W for M81 Dwarf A, and F606W and F814W for UGC8760. have DOLPHOT flags of 0 or 2 in those filters, and which fall within the region of full-filter coverage for that target. The stellar number density is computed in the reference image filter of each target over a magnitude range, typically between roughly 20^m–27^m, where the completeness is expected to be close to 100%.¹⁰¹⁰10Since this magnitude range has to be determined before performing actual ASTs, we rely on our team’s previous experience with similar HST data as well as the results of previous relevant studies (e.g., Dalcanton et al., 2009; Choi et al., 2020a). We confirmed the chosen range for each target was reasonable using the results of the full suite of ASTs. We utilize 5–10 linear bins in stellar number density. The number of bins is dependent on the dynamic range of each galaxy, with 5 (10) stellar density bins used for a galaxy with the smallest (largest) dynamic range in stellar density.

We randomly select model SEDs from the dust-applied stellar physics grid until at least 70 models are selected per flux bin and per stellar number density bin, in each filter. This first suite of input ASTs comprises roughly half of the final input ASTs of each galaxy, and is designed to cover the entire physics model grid (including areas well removed from the observed CMD space), which is required for the BEAST stellar SED fitting.

To ensure adequate ASTs to support CMD modeling, we supplement the above with a higher density of ASTs within the region of observed CMD space. This supplement includes at least 4,000 (and up to 20,000) model SEDs per stellar density bin, focusing on the observed color and magnitude ranges for a given target. This ensures sufficient ASTs to compute completeness as a function of color and magnitude, which is essential for the CMD modeling.

Figure 9 shows the CMD location of the DDO210 input ASTs which cover the extinguished physics grid (top) and the supplemental ASTs which include a higher density of stars in the region of the observed CMD (bottom). The number of flux bins, number of model SEDs per flux bin, number of stellar number density bins, and number of supplementing model SEDs per spatial density bin vary with the properties of the individual galaxies (e.g., dynamic range of stellar number density) and observations (e.g., depth). These numbers are empirically determined based on extensive tests and previous studies (e.g., Gordon et al., 2016; Choi et al., 2020a; Van De Putte et al., 2020; Yanchulova Merica-Jones et al., 2021). In general, this results in an input AST list that is a factor of $\sim$1.2 – 2$\times$ greater than the number of sources observed with a signal-to-noise ratio $>4$ in at least one filter.

The input AST list for each target, generated as described above, consists of assigned locations and magnitudes for each of the filters included in the target’s dataset. Each of the entries in the input AST list is used to inject an artificial star into the observed images, and the same photometric measurement routine is run as described in Section 3.2. The artificial stars are injected one at a time, so that the measurement of a given artificial star is not affected by the others in the input AST list. This procedure results in the same photometric measurements and quality metrics that are reported for the observed astronomical sources (Section 3.3).

For the following analysis, an artificial star is deemed to be “recovered” in a given filter if the absolute difference between the input and measured magnitudes is less than 0.75 magnitudes, the star is recovered within two reference pixels of its input location, and the photometric measurement of the star passes all GST quality criteria as described in Section 3.3. ASTs placed in areas affected by diffraction spikes are not included in the computation of the recovery fraction, bias, or photometric uncertainty as a function of magnitude.

For each target, we compare the DOLPHOT measurements of the injected ASTs to the input magnitudes to evaluate the recovery fraction of ASTs (Figure 10, Table 7), as well as the accuracy (Figure 11) and uncertainty (Figure 12) of the measured magnitudes, as a function of magnitude in each filter. As discussed in Section 5.1, we limit this analysis to the spatial region with full-filter coverage, with the exception of WLM Position 1 (Section 6).

Significant spatial variations in the photometric quality are to be expected within a given filter. These variations may be due to intrinsic changes in the stellar density within a galaxy, leading to varying levels of crowding impacts on the stellar measurements. However, given the heterogenous nature of the archival observations, as well as the presence of chip gaps in the ACS, WFC3/UVIS, and WFPC2 imaging, spatial variations in the photometric quality can also be due to any combination of the following: observations obtained in the same filter but with different exposure times; partially overlapping observations in the same filter resulting in more total exposure time; or areas falling within a chip gap in a filter, resulting in less total exposure time. Thus, the AST-based analyses that follow should be considered to be representative of the general quality of the LUVIT data for a given target.

As previously cautioned, all of the AST-based results reported above should be viewed as representative values, computed over the spatial region with full-filter coverage for each target. Significant spatial variations in completeness, photometric uncertainties, and bias as a function of magnitude in the LUVIT targets can be caused by either the heterogeneity of the archival datasets and/or a significant range of observed stellar densities encountered across the field of view of the LUVIT observations, as discussed in Section 5.4. Moreover, although the LUVIT targets are low-mass dwarf galaxies, the blending of sources does appear to impact the measurements in some filters in many of the LUVIT targets, as demonstrated in Figure 12, although the effect is minor for most targets, especially compared to the impacts seen in more massive galaxies (e.g., Williams et al., 2014). In this section, we explore the impact of observed stellar surface density on the completeness limits and measurement bias.

We use 5″ by 5″ spatial surface density maps, computed analogously to those computed for creating the input AST lists (Section 5.2), but restricted to sources passing all stellar quality criteria (Section 3.3) and within the full-filter area. For each target, we compute the 25th, 50th, and 75th percentiles of the distribution of stellar spatial densities in the reference image filter, creating four bins of stellar density for each target. The upper limit of the lowest density bin is defined by the source density $\Sigma_{25}$ at which 25% of the observed sources fall in a 5″ by 5″ source density map “pixel” with source density $\Sigma\leq\Sigma_{25}$. The upper limit of the second-lowest density bin is then defined by the source density $\Sigma_{50}$ at which 50% of the observed sources fall in a region with source density $\Sigma\leq\Sigma_{50}$, with $\Sigma_{25}$ forming the lower bound on the second-lowest density bin. The third and fourth density bins are defined in the same manner. Thus, the limits of each bin are tied to the specific dynamic range and stellar density distribution of the observations (within the full-filter area). Each AST is then assigned to a stellar density bin based on the observed stellar density $\Sigma$ (again using the 5″ by 5″ spatial density map) at the spatial position of the injected AST. We then compute, for each stellar density bin and for each filter, the fraction of recovered ASTs as well as the photometric bias. This method ties the analysis of all filters to the stellar density distribution of the stellar sources detected in the reference image filter.

The results are summarized for all targets in Figure 13, which shows the difference in the 50% (top panel, $\Delta m_{50\%}$) and 80% (bottom panel, $\Delta m_{80\%}$) completeness limits between the lowest and highest spatial density bins for each filter in each target. We choose a threshold of $\pm 0.2$ magnitudes in $\Delta m$ to identify potential crowding impacts. Figure 14 displays the recovery fraction and bias as a function of magnitude (input and recovered, respectively) in each of the four spatial density bins for all targets that are identified as being potentially affected by crowding in at least one filter in Figure 13 (18 of the 23 LUVIT pointings). For differences of $|\Delta m|<0.2$, the difference in the recovery fraction curves and the bias curves are small. This can be seen by comparing, within a given target, the recovery fraction and bias curves of the lowest and highest stellar density bins for filters with $|\Delta m|<0.2$ in Figure 14.

A few general patterns are apparent across the LUVIT pointings. The NUV observations show no significant impact from crowding on either the limiting magnitudes of the observations or bias in the recovered magnitudes. However, most LUVIT pointings reach sufficient observed stellar densities in the optical bands to impact the depth of the stellar catalogs in the higher density regions. A number of the LUVIT pointings also show impacts from stellar crowding in the NIR bands. These impacts are typically larger in F110W than in F160W, which is likely due to the fact that the F110W observations are also typically $\gtrsim 1$ magnitude deeper than the F160W observations. For the targets and filters included in Figure 14, the completeness limits are generally larger, and the absolute magnitude of the bias (particularly the 16th- and 84th-percentiles of the distribution) generally smaller, in the lower spatial density bins than in the higher spatial density bins.

The two galaxies with the largest spread of $\Delta m$ values are NGC 4163 and UGC 9240. These two galaxies have a large number of stars passing our quality criteria (Section 3.3) in the optical bands combined with some of the highest central stellar surface densities in the LUVIT sample, with the large majority of the stars concentrated within a portion of one HST/ACS chip (Figure 4). In contrast, the WLM-POS2 LUVIT target has the highest number of stars passing our quality criteria in the optical of any LUVIT pointing, but they are much more evenly spread across the HST/ACS field of view. The very large $\Delta m_{50\%}$ seen in F555W and F814W in Sextans A is due to the significant heterogeneity in the depth of the archival observations used in the LUVIT reductions (Section 6).

For a small number of targets, the full-filter region misses areas of higher stellar density which fall outside the full-filter region. For most of these targets (WLM-POS2, UGC8651, UGC8833, and UGCA292), these are relatively isolated, small areas of higher density. Relative to these targets, WLM-POS1 has a significantly larger share of area at higher stellar density that falls outside the full-filter region, while the highest stellar density areas in Sextans A and Sextans B fall outside the full-filter region.

Finally, we remind the reader that a few of the LUVIT targets have spatial variation in observed depth for a given archival filter due to the inclusion of multiple pointings in that filter with different depths (Sextans A, noted above, is an extreme example). For these targets, some portion, or even the majority, of the differences in the recovery fraction and bias as a function of stellar density may be due to correlations with the observational heterogeneity and not to the impacts of stellar crowding.

In summary, the impact of blended sources on the photometry in higher (observed) stellar density regions is relatively small for most of the filters in most of the LUVIT stellar catalogs. However, there are a number of targets where the recovery fraction and bias differ significantly with sky location for some subset of the filters. For these targets, any analyses that extend to magnitudes at which the recovery fraction is $<1$ should take care to understand the potential impact on their results of the differences in recovery fraction and measurement bias as a function of spatial position.

While optical CMD fitting provides a reliable SFH across cosmic time (e.g., Gallart et al., 2005; Tolstoy et al., 2009), the LUVIT data demonstrate that the NUV CMDs have the potential to better constrain both the younger populations which dominate the UV flux, and the dust properties around the younger stellar populations. We are quantifying the extent to which integrating the NUV CMDs with the optical CMDs enhances the precision and accuracy of recent SFH determinations. Using LUVIT NUV and optical stellar photometry, we can quantify the impact of including NUV data on measuring the SFH by simultaneously modeling the NUV and optical CMDs for the first time (Y. Choi et al., in prep).

A primary motivation for obtaining NUV through NIR imaging in the LUVIT galaxies was to enable SED fitting of resolved stars, thereby constraining both stellar [e.g., age, mass, and metallicity] and line-of-sight dust extinction parameters [dust column ($A_{V}$), grain size distribution ($R_{V}$), and the amount of mixing between the Milky Way- and Small Magellanic Cloud-type extinction curves ($f_{A}$)] in the LUVIT galaxies.

In addition to enabling the creation of spatially resolved maps of the mean stellar and dust extinction parameters in LUVIT galaxies, such analyses can also be used to infer the intrinsic flux being output by the stars in a galaxy, including at wavelengths not included in the original observations. For example, Choi et al. (2020a) used the BEAST to perform SED fits on individual stars in a similarly broad panchromatic dataset in NGC 4214. The SED fits were used to infer the total ionizing flux being output by the stars in that galaxy, resulting in observational constraints on the spatially resolved, local escape fraction of ionizing photons and yielding a more accurate, and considerably higher than previously found, estimate of the global escape fraction for this high-$z$ analog (Choi et al., 2020a).

The NUV-NIR CMD locations and SED fitting has also been used to identify and characterize massive stars in nearby galaxies. For example, Lindberg et al. (2024) used BEAST fits to identify over 42 thousand massive star candidates across the PHAT region of M31 (Dalcanton et al., 2012a). In Leo A, Gull et al. (2022) used the LUVIT stellar catalog to identify massive star candidates for spectroscopic observations, prioritizing the brightest stars in the F336W vs. (F336W – F475W) CMD. Using the BEAST, Gull et al. (2022) compared stellar parameters derived from SED fitting of the LUVIT photometry of these massive star candidates with parameters derived by fitting the stellar spectra (specifically, comparing the effective stellar temperature, surface gravity, and stellar luminosity). They found remarkably good agreement in these parameters for main-sequence (MS) stars with satisfactory fits in both methods, indicating that NUV through NIR photometric SED fitting is promising for identifying and characterizing the most massive stars in low-metallicity galaxies too distant in which to obtain resolved stellar spectra. They also found that B-type stars with emission lines stand out as outliers in the SED fits, and are well distinguished from massive MS stars when comparing the F275W vs. (F275W – F475W) CMD to the F475W vs. (F475W – F814W) CMD.