The JWST Early Release Observations

The NIRCam image centers module B on the cluster, and module A on an adjacent off-field. While the ERO product featured only the cluster module, the off-field module represents an additional, serendipitous deep field. We obtained NIRCam images through six broadband NIRCam filters, spanning F090W through F444W (see Table 1). The exposure time was designed to achieve a point-source sensitivity of AB ∼ 29.8 mag, as estimated using the JWST Exposure Time Calculator (5σ), aimed at matching the depth of the Hubble Ultra Deep Field as observed with WFC3/IR (HUDF12/XDF; Ellis et al. 2013; Illingworth et al. 2013; Koekemoer et al. 2013), and surpassing by at least a factor of 10 the depth of Spitzer IRAC channel 1+2 imaging for similar fields. In practice, due to improved NIRCam sensitivity in these filters by up to 30%, as uncovered during commissioning (Rigby et al. 2022), combined with the lower Zodiacal background of this field compared with the HUDF, the ultimate depth of the SMACS J0723.3-7327 field may be somewhat deeper, perhaps reaching or surpassing AB ∼ 30 mag over much of the field.

We used the INTRAMODULEX dither with nine dither positions to optimize image quality and efficiently remove cosmic rays and bad pixels, and the MEDIUM8 readout pattern to minimize detector read noise. Given the moderately low galactic latitude of the field at −237 it does contain a number of bright Milky Way stars.

We used MIRI to image the cluster in four filters (Table 1). We prioritize depth over coverage, and use a single MIRI tile centered on the cluster, reaching 5σ depths of AB ∼ 26.3 mag at F770W, ∼25.1 mag at F1000W, ∼24 mag at F1500W, and ∼23 mag at F1800W. We use relatively long ramps with FASTR1 of 100 groups to minimize read noise, except for F1800W, which uses a 50 group ramp given the higher background in this wave band. All images use a 10 point medium cycling dither.

We obtained slitless grism spectroscopy with NIRISS of a single tile centered on the cluster, using two different filters to increase the chances of identifying some galaxies that show interesting emission features at high redshift. The F115W and F200W filters are able to detect emission-line galaxies via Hα at z = 1.0–2.4 and Hβ along with the [O iii] 4959+5007 Å doublet at z ∼ 1.1–3.6. We used both orthogonal gratings to minimize the effects of contamination in the crowded cluster field. Using an eight-point medium dither pattern, and a total exposure time of 2.8 ks, we reach SNR ∼ 10–30 per channel on the continuum for an AB ∼ 23 mag galaxy, but are also able to detect much fainter emission-line galaxies.

The NIRSpec Multi-Shutter Array (MSA) was used in its medium-resolving configuration to investigate selected galaxies in greater detail, and with wider spectral coverage, reaching 5 μm. The medium resolution grating balances sensitivity, velocity resolution, and the number of spectra that can be fitted on the NIRSpec detectors. The G235M and G395M gratings include Hα at z = 1.7–6.9, and the easily identifiable trio of the Hβ line and [O iii] 4959+5007 Å doublet out to z ∼ 9. We use a three-shutter slitlet with a matching three-dither pattern, using the NRSIRS2 readout pattern to limit 1/f noise.

The NIRCam six-band images, obtained as part of this program, were to derive a catalog and photometric redshifts using the Eazy package (Brammer et al. 2008). We used an astrometric solution referenced to Gaia, and supplemented by the RELICS HST/ACS imaging in the area outside the NIRCam footprint (Coe et al. 2019). We select galaxies in the field out to z ∼ 10 with AB <26 mag. For the handful of z > 6 galaxy candidates we included objects with AB ∼ 27 mag, anticipating that strong emission lines will still easily be detectable. Empty shutters were included in the configuration for subtracting the background.

Among a total set of 41 primary and four secondary targets, we included the four highest redshift objects at z ∼ 6–10, the giant gravitationally lensed arcs, galaxies at z ∼ 1–6, and cluster galaxies at z ∼ 0.4.

The NIRCam image of Stephan's Quintet is the largest mosaic of the ERO release, consisting of three pairs of dithered tiles, arranged such that each pair has a 71.5% overlap in the direction along the two modules. In total, the final product includes 153.5 Mpixels. The double-tile strategy was designed to maximize the uniformity of the depth across the image to support the production of even-looking color images, and to maintain a minimum of five dithers on any given location to be able to produce images free of cosmic rays and bad pixels. Further, we used the FULLBOX dither pattern to image a rectangular field, while minimizing the need to crop irregular edges.

Since the redshift of the merger is large enough that narrowband filters will generally not contain their target lines, wide filters are used, spanning a wide range in wavelengths from 0.9 to 4.5 μm (see Table 2). The shortest band includes structure from dark dust lanes within each galaxy, while the longest band enhances the field of background galaxies, and includes emission from the 3.3 μm PAH feature and the strong rotational S(9) line from H₂. The exposure parameters were chosen to maximize dynamic range (high sensitivity, but without saturating too many brighter sources), including the brightest areas around the NGC 7319 AGN core. The NIRCam images reach 5σ depths of AB ≳ 29 mag in the deepest areas of the field.

Since MIRI covers less than a quarter of the field of view of a single NIRCam tile, there was insufficient time to cover the full NIRCam area. Consequently, the MIRI image covers the central galaxies, NGC 7318, NGC 7319, and NGC 7320, using four tiles. The two tiles centered on NGC 7320 were observed significantly later than the tiles on NGC 7318 and NGC 7319, and consequently the relative aperture position angles differ by a few degrees. The MIRI image consists of three filters, F770W, F1000W, and F1500W, intended to trace PAH emission, as well as to reveal embedded star-forming clusters. The pointing including the NGC 7319 core was a critical inclusion, needed for context with NIRSpec IFU and MIRI MRS observations of the Seyfert 2 AGN. We used eight dither points of the large cycling pattern to maximize the spatial coverage of a single tile.

We obtained a single NIRSpec IFU tile centered on the Seyfert 2 AGN in NGC 7319. We use the PRISM for maximal sensitivity and instantaneous spectral range (0.7–5.2 μm). It is noteworthy that the redshift of NGC 7319, albeit relatively small, allows the rest-frame optical Hα line to be included in the NIRSpec coverage. Sensitivity at shorter wavelengths was a particular consideration, given how red (and likely embedded) the AGN is. At the same time, the IFU exposure parameters avoid saturation at the longer wavelengths. eight dither points from the large cycling pattern maximize the spatial resolution. The single tile covers the central parts of the AGN outflow/jet structure.

A single MIRI Medium Resolution Spectroscopy (MRS) tile was also observed, centered on the same region of the NGC 7319 core that was covered by the NIRSpec IFU. We observed all three sub-bands for full spectral coverage from 4.9 to 28 μm. The MIRI MRS observation was supported by a background observation, centered on a blank part of the sky off the galaxy group, to help remove the telescope background. Given the brightness of the AGN, the MIRI spectrum generally has a very high signal-to-noise (>100) across the covered spectral range.

A key objective of the Carina image was to demonstrate, not only how the infrared range can be used to reveal embedded young stars, but also how wavelengths beyond ∼2 μm trace bright molecular emission from particularly PAHs and H₂, leading to spectacular, high-resolution vistas.

To this end, we observed four mosaic tiles with NIRCam, using a pairwise overlap of 71.5% in the horizontal direction in order to get as much uniformity in the depth as possible. This is the same strategy as that employed for Stephan's Quintet.

We used the FULLBOX five-point dither pattern (10 dither points with the overlapping tiles) to be able to construct a rectangular image. For use with color images of galactic star-forming regions, we designed the following six-filter set: F090W and F200W to image scattered light from dust and show extinction colors of the background stellar field, while optimizing spatial resolution and sensitivity at the shortest wavelengths. The F187N filter captures ionized gas via the bright Paα, F470N images H₂ from embedded jets and outflows, F335M images emission from the 3.3 μm PAH band, and F444W provides a long-wave sensitive point to dust scattering from large grains. Exposure times are set to not saturate on extended nebular emission, although many point sources will inevitably saturate in the broad filters (see Table 3). In particular the F335M filter is critical in bringing out the highly textured structure from the surface of the molecular cloud; the PAH emission traces a combination of cloud density and the strength of the local ultraviolet field. Together, these aspects create a thin layer tracing a surface through the molecular cloud. Since the 3.3 μm feature is the shortest wavelength PAH feature, this filter provides the highest resolution view of the molecular boundary layer of the photodissociation region available to JWST.

As in the case of Stephan's Quintet, the field is too large to completely cover with MIRI imaging. However, the MIRI image mosaic does use five tiles to cover a significant fraction of the interface region of the Carina bubble, overlapping with the NIRCam field of view, and including several protostellar associations and associated outflows. An eight-point cycling large dither pattern was used to provide good depth coverage across the panorama. We used four filters relevant to star-forming regions: F770W and F1130W for PAH emission, F1280W to include emission from the 12.81 μm [Ne ii] line, and F1800W for a long-wavelength point tracing cooler dust and the presence of thick protoplanetary disks around the young stars in the field. Exposure times are set to not saturate the nebular emission, but some bright point sources inevitably saturate (see Table 3).

We imaged the nebula with a combination of broadband and narrowband filters, targeting known spectral features: F090W for the background galactic field and scattered light on dust. F356W was used to detect any PAH emission and to add colors of background galaxies in combination with F090W. F187N and F405N trace the bright Paα and Brα hydrogen recombination lines, respectively. Finally, F212N and F470N image the strong rovibrational and rotational H₂ emission lines in the nebula, tracing different excitation temperatures. The narrowband filter exposure times are set to reach high signal-to-noise while not saturating, to increase dynamic range, and to reveal turbulent, small-scale structures, particularly in the H₂ emission from the outer parts of the nebula. The exposure time estimates were informed by the surface brightness as measured in archival Spitzer IRAC1+2 images (Figure 4). The nebula is fully contained within NIRCam module B, and we use the INTRAMODULEX dither pattern with eight pointings to get uniform coverage.

Zoom In Zoom Out Reset image size

We imaged with MIRI NGC 3132 using two tiles to cover roughly the same field of view as NIRCam Module B. The exposure times are set using archival Spitzer IRAC and MIPS images, along with Spitzer spectroscopy to model performance, as the mid-infrared nebula is entirely dominated by emission lines. We target the 11.3 micron PAH specifically, as it is seen in the Spitzer spectrum. We also target F1280W to measure the known bright [Ne ii] line, and the F1800W filter to image the 18.71 μm [S iii] line, as well as any cool dust emission. We use the eight-point cycling dither pattern, and two mosaic tiles to roughly cover the NIRCam module B field, albeit at a slightly different position angle.

One of the characteristics of the NIRCam detectors is the presence of low-level electronic noise that is highly correlated in time, across output amplifiers and reference pixels, referred to as "1/f noise" (Moseley et al. 2010). This is manifested as linear stripes, which change gradually in their amplitude across the detector. This electronic effect is at a relatively low level and did not significantly impact the NIRCam images of NGC 3324, NGC 3132, or the bright galaxies in Stephan's Quintet. However, it was noticeable in fainter regions of the images, and especially so for the deep imaging of SMACS J0723.3-7327. The version of the pipeline available at the time of the ERO observations did not include a correction for this, so a simple correction was applied to each exposure. Specifically, we measured the median value across each row, after masking bright sources, and then adjusted the values in each row accordingly. While some additional higher-order variations might still be present, this procedure provided a first-order correction to improve the low-level background quality and noise properties of the resulting mosaics.

Another aspect of the NIRCam detectors is the presence of an electronic bias level that is removed during calibration by applying a "superbias" reference file. While this process is generally sufficient to remove this electronic bias, small residual offsets in the background level may remain after calibration, and these may be different for each detector. The JWST pipeline includes a "skymatch" step to match and remove any such residual background, and there were no visible residuals remaining for the ERO observations of NGC 3324, NGC 3132, or Stephan's Quintet. However, for the deep image of SMACS J0723.3-7327, some low-level residual background differences remained, visible as sky-level offsets between different detectors. Consequently, an additional iteration of background subtraction was carried out by masking bright sources and extending those masks to also exclude fainter emission before remeasuring the background levels. This successfully corrected the remaining residual background differences in these images.

The NIRCam instrument is composed of two independent modules (A and B), each of which has five separate detectors, four of which are in the Short Wavelength Channel in a 2 × 2 grid, with the remaining detector being in the Long Wavelength Channel. For all 10 of these detectors, measurements of their relative positions were available from ground-test data. While preliminary on-orbit measurements had been obtained to refine the detector astrometry, not all of these were available at the time of the ERO observations. Initial calibrations revealed inaccuracies up to ∼02 in the relative locations of some of the detectors, as well as about +/−05 in their relative orientations, compared to ground-test data. Therefore, for all the NIRCam images, these offsets had to be corrected by invoking a part of the pipeline that solves for offsets ("tweakreg") in a custom setting, essentially registering each of the 10 detectors independently to an external catalog of the field, solving for their shifts and rotations separately for each filter. For the NIRCam imaging of NGC 3324, NGC 3132, and Stephan's Quintet, the external catalog consisted of Gaia-DR3. ¹⁴ For the images of SMACS J0723.3-7327, too few Gaia stars were available to provide a robust direct alignment, so the observations had to be aligned to RELICS HST observations of the field, which also needed to be aligned to Gaia-DR3. This successfully corrected all the astrometric errors, achieving a relative precision of ∼5 mas and absolute astrometric accuracy of ∼10 mas, and enabled all the images to be precisely aligned and combined into final mosaics.

The relative astrometry between dithered images taken using the same guide star is generally accurate to a few mas (Rigby et al. 2022)—more than sufficient for MIRI imaging with a pitch of 110 mas. However, the relative astrometry between mosaic tiles observed using different guide stars can have much larger offsets, driven by the precision of the guide star catalog. Generally, offsets between different guide stars of a few 01 are common, with occasional differences as high as 1''–2''. While this is enough to cause noticeable registration problems with MIRI imaging mosaics, offsets are usually straightforward to measure and correct using the "tweakreg" step in the calwebb_image3 pipeline step, provided that there are a sufficient number of point sources within each tile. This is not always the case for MIRI imaging, especially at longer MIRI wavelengths where stars become very faint.

We refined the MIRI imaging astrometry and registration by deriving offsets between dithers and tiles using "tweakreg" on the shortest wavelength band (e.g., F770W), and then applying the same offsets to every other longer-wavelength band. Further, the offsets in (V2,V3) reported by "tweakreg" for each F770W exposure were averaged over all the dithers within a mosaic tile for enhanced precision and robustness against a lack of point sources in each individual dither. This approach takes advantage of the excellent pointing stability and accuracy of the JWST platform. Finally, the "tweakreg" step registers the absolute astrometry to the Gaia frame of reference.

For observations of faint, low-background targets, the residual instrumental structure becomes significant. This was corrected by subtracting an average background image constructed by sigma clipping the stack of all exposures in detector coordinates to remove sidereal sources. While this step significantly reduced the residual instrumental background structure, it did imprint weak, large spatial scale structures due to imperfectly removed astrophysical extended emission.

For sources that are very bright compared to the background emission, detector artifacts causing bars to appear along detector columns and rows appear. We used a simple correction by constructing column or row median profiles, and subtracting them from the appropriate detector regions in each exposure. The sources and regions suffering from these effects were identified visually in the mosaics. The column/row profiles were constructed and applied independently above/below and left/right of bright sources.

The NIRISS/SOSS images were processed using version 1.5.2 of the JWST pipeline. We worked with the Stage 1 products to produce the transmission spectrum, i.e., with the rates per integration. From there, the 2D spectra were analyzed and extracted using custom routines specialized to deal with NIRISS/SOSS data. ¹⁸

First, we traced the Order 1 and Order 2 spectra using a custom routine that performs a cross-correlation of each column in the image with a sum of two Gaussians separated by 15 pixels. Once the position of the center of those traces was obtained, we fitted those with splines in order to smooth their shapes. Next, we performed 1/f noise corrections by subtracting the median of all integrations from each individual integration. This left us with a frame for each integration containing detector-level effects. With these frames, we obtained the level of 1/f noise by computing the median number of counts within 20 to 35 pixels from the trace. These counts were then subtracted to each pixel within a 15 pixel radius from each trace of the original integrations.

The final step of our detector-level data reduction removed the zodiacal background from each integration. Under the assumption that the zodiacal background stays constant throughout the exposure (an assumption we tested with our data), we scaled the background model provided in the JWST Documentation (JWST 2016) using the median of all the integrations in the exposure. In particular, we used columns 500 through 800 and rows 210 through 250 of both the data and the model, taking the ratio of both to find the scaling factor. We used the second quartile of this ratio to estimate the scaling factor between the data and the model, which gave a ratio of about 0.47. This value was then used to remove the zodiacal background signal from each individual integration. The spectra were extracted using simple box extraction with an aperture radius of 15 pixels.

The 280 spectra implied 280 fluxes measured in each wavelength bin (2044 for Order 1 and 551 for Order 2), which formed our set of 2591 transit lightcurves to analyze. Each lightcurve was divided by the out-of-transit median flux and then fitted using juliet (Espinoza et al. 2019). These were fitted using a batman model (Kreidberg 2015) with the best-fit orbital parameters obtained from an analysis made on the white-light lightcurves fixed in the wavelength-dependent analysis. The limb-darkening coefficients were fitted at each wavelength using the sampling scheme of Kipping (2013) and a square-root law, but using limb-darkening coefficients obtained through the limb-darkening software combined with the SPAM algorithm as priors (Espinoza & Jordán 2015). The priors were set on the transformed coefficients (q₁, q₂), with truncated normal distributions centered on these theoretical predictions, having standard deviations of 0.1 and constrained to be between 0 and 1. In addition to the transit model, a Gaussian Process was used to model any trends and/or instrumental systematics. In particular, a Matèrn 3/2 kernel in time was used, which defined two extra parameters for the fit: the amplitude of the Gaussian Process and the timescale of the process. Additional free parameters were a mean out-of-transit flux, a photometric "jitter" term added in quadrature to the error bars for each wavelength, and the planet-to-star radius ratio. We used the dynamic nested sampling algorithm implemented in the dynesty package as described in Speagle (2020) to sample from the posterior distribution of the parameters. While the posterior distribution for all parameters was obtained for our analysis, the only observable presented in the ERO package was the square of the planet-to-star radius ratio—i.e., the transit depth as a function of wavelength. While the transit depths were calculated at the resolution level of the instrument, we presented a spectrum binned to R = 100.