INSTRUMENT SCIENCE REPORTS (ISRs)

An LED provides WFC3/UVIS with the capability to increase background levels with user-directed flashes that can help to improve charge transfer efficiency (CTE) in images with low backgrounds. Analyzing data from 2012-2021, we completed several tests to check for changes in power and illumination pattern of the LED onto the UVIS detector over time. By studying the subarray and full-frame post-flashed dark images over time we find there is a decrease in the measured post-flash mean over time. This change is in line with previously observed changes in WFC3 sensitivity for filters with similar spectral properties as the LED, measured to be about 0.2% per year (Calamida et al. 2021, Marinelli et al. 2022). We interpret this to indicate that there is no systematic decay in the LED illumination that cannot be attributed to previously reported detector sensitivity changes. Following that result, we investigated whether there was any time dependence in the post-flash reference files if binned by yearly or biyearly cadences. We used these newly created time-dependent reference files to produce FLC images for various GO science data and determined that the time-dependent post-flash reference files do improve the quality of the science data without significantly affecting the measured noise. We report that the average mean measurement of our science images change up to 2.64% with the time-dependent reference files while the standard deviation change is measured to be less than 1/1000 of a percent. We determined that using yearly reference files was the most accurate calibration available as it balances the number of input files, which affects the signal-to-noise, and the time covered by the input data, which affects the change in measured post-flash. We therefore delivered new yearly post-flash reference files to the WFC3/UVIS data calibration pipeline calwf3. They have been available as of December 2022, and data requested from MAST since their delivery is post-flash-corrected with the new reference files.

We present the STScI WFC3 project webpage, Transiting Exoplanets List of Space Telescope Spectroscopy, TrExoLiSTS. It tabulates existing observations of transiting exoplanet atmospheres, available in the MAST archive made with HST WFC3 using the stare or spatial scan mode. A parallel page is available for all instruments aboard JWST using the spectral Time Series Observation (TSO) mode. The webpages include observations obtained during primary transits, secondary eclipses and phase curves. TREXOLISTS facilitates proposal preparation for programs that are highly-complementary to existing programs in terms of targets, wavelength coverage, as well as reduces duplication and redundant effort. Reference for the quality of the HST WFC3 visits taken more than 1.5 years ago are made available via including diagrams of the direct image, white light curve and drift of the spectral time series across the detector. Future improvements to the webpage will include: Expanding program query to other HST instruments and reference for the quality of JWST visits.

We present new WFC3 UVIS CCD absolute gain measurements for December 2018 (Cycle 26) through June 2022 (Cycle 29). We employ the mean-variance technique using internal flat field image pairs taken in unbinned, 2×2, and 3×3 binned modes. As in previous programs, there are two epochs of data per Cycle, one in December and a second in June the following year. We find that the final gain values for the latest June 2022 unbinned data are 1.587 ± 0.007, 1.578 ± 0.004, 1.607 ± 0.005, 1.592 ± 0.006 e-/DN for amps A, B, C, and D respectively. In addition to analyzing the newest Cycles, we also looked at the full evolution of gain measurements over time. We find that while the gain measurement continues to increase Cycle to Cycle, it remains stable and consistent within 1-2% of past Cycles and the initial TV3 (ground) testing. We also investigate charge transfer efficiency (CTE) effects, and determine that increasing gain measurements is partially attributed to CTE degradation.

The infrared channel of Wide Field Camera 3 (WFC3) on the Hubble Space Telescope (HST) is frequently used to obtain precision photometric measurements. We investigate the change in the sensitivity of the IR channel over time by comparing photometry of outer, less crowded regions of stellar cluster images at multiple epochs. The channel appears to be losing sensitivity at a rate of approximately 0.13±.02% per year with no apparent wavelength dependence, though the slope and observation dates vary from cluster to cluster. Staring mode observations of the standard stars appear to show more inconsistent results, even within the same filter, but this is likely due to the presence of systematics such as detector preconditioning from prior observations.

As of late-2019, MAST data products for ACS and WFC3 include improved absolute astrometry in the image header World Coordinate System (WCS). The updated WCS solutions are computed during pipeline processing by aligning sources in the HST images to a select set of reference catalogs (e.g. Gaia eDR3). We compute statistics on the alignment fraction for each detector and estimate the uncertainties in the WCS solutions when aligning to different reference catalogs. We describe two new types of Hubble Advanced Products (HAP), referred to as Single Visit Mosaics (SVMs) and Multi Visit Mosaics (MVM), which began production in MAST in late-2020 and mid-2022, respectively. The SVM products include an additional relative alignment across filters in a visit, and the drizzled images are used to generate point source and segment catalogs during pipeline processing. These catalogs supersede those produced by the Hubble Legacy Archive and will be the basis of the next version of the Hubble Source Catalog. The MVM data products combine all ACS/WFC, WFC3/UVIS, or WFC3/IR images falling within a pre-defined 0.2° x 0.2° 'sky cell' for each detector+filter, which are drizzled to a common all-sky pixel grid. When combining observations over a large date range, MVMs may have photometric errors of several percent or systematic alignment errors when combining visits with different catalog solutions. We therefore recommend these to be used as ‘discovery images’ for comparing observations in different detectors and passbands and not for precise photometry.

Using five years of observations from the Hubble Space Telescope (HST) Wide Field Camera 3 (WFC3), we assess the changing sensitivity rate of the two WFC3/UVIS charge-coupled devices (CCDs) and evaluate the photometric repeatability of the spatial scan observing mode in comparison to the standard staring mode. We perform aperture photometry on vertical, linear spatial scans of two white dwarf standard stars (GD153 and GRW+70D5824) taken on corner subarrays of the WFC3/UVIS detector, and compute the rate of photometric sensitivity decline from 2017 to 2021. To gauge the relative accuracy of scans, we compare sensitivity losses for staring mode observations over the same 5-year time scale and those acquired over longer time scales. After removing the time-dependence of the relative photometry, dispersion of the residuals is used as a proxy to measure repeatability of observing modes, and thus assess precision. We establish that spatial scans are more precise than staring mode observations. Scans with UVIS 1 show 2.4 times less residual noise than their staring mode counterparts; for UVIS 2, residual noise for scans is 2.5 times less than residual noise for staring mode. For scans, sensitivity losses are relatively flat independent of wavelength on both UVIS CCDs, with no evidence of contamination. UVIS 2 appears to have slightly higher losses (-0.17 +/- 0.01 %/yr) compared to UVIS 1 (-0.12 +/- 0.01 %/yr). When measured over the same time period, spatial scans and staring mode observations yield filter-dependent loss rates that agree well with each other in most filters within computed uncertainties.

This document describes and announces the public release of a software routine, hst1pass, that has been optimized for PSF-fitting photometry on undersampled HST images. Previous versions of the code have been written for individual HST instruments and released as a part of various instrument-specific ISRs. But over the last few years, the code has been generalized to work for all of HST’s main imagers (WFPC2, ACS/HRC, ACS/WFC, WFC3/UVIS and WFC3/IR). It also runs in aperture-photometry mode and, as such, can be run on _drz and _drc products or even non-HST images. The program itself is written in FORTRAN, but a simplified version in Python will be made available soon. In its typical usage, the user specifies some simple finding parameters, and the routine reads in (1) an HST image (_flt or _flc), (2) a PSF and, (3) a distortion solution. The routine then goes through the image pixel-by-pixel and returns a list of stars found and measured in the image. This star list can then be collated with similar lists, such as from a set of dithered exposures from the same program. In a future ISR the collation process will be described in more detail, but a simplified version of the collation software is provided here to facilitate preliminary analysis.

"Figure-8 ghosts" are a type of HST/WFC3 anomaly caused by light reflecting off a part of the WFC3/UVIS detector. Since anomalies affect both science and calibration data, it is vital that we flag them to properly support the WFC3 user community. Traditional methods of anomaly detection will become impractical as the volume of data sets increases in future missions such as the Roman Space Telescope. WFC3 data provides the opportunity to test anomaly detection with machine learning in current generation astronomical images. We trained five convolutional neural networks (CNNs) of varying architectures to classify figure-8 ghosts in WFC3/UVIS images. This report outlines the data selection, preparation, and augmentation to construct our training, validation, and test sets. Our models' accuracies range from 62% to 83%, showing that CNNs can be a powerful tool to detect other anomalies in WFC3 or other images. In addition to applying machine learning to a specific astronomical task and discussing future work, we consider astronomy's role in machine learning, i.e., innovating computer vision with large images, as a potential challenge for the future.

We compute new UVIS encircled energy (EE) curves from images of HST standard stars observed from 2009-2020, after correcting for temporal changes in the sensitivity of each CCD and filter. The latest UVIS photometric calibration (Calamida et al. 2021) makes use of the updated EE values presented in this report for two filters (F275W and F814W). We extend this analysis to five additional filters (F336W, F200LP, F350LP, F775W, and F850LP) to investigate differences between the 2009 EE, derived just after WFC3’s installation, and the 2016 chip-dependent EE, computed by averaging inflight observations over ~6 years. To recompute the EE, we rescale the calibrated FLC science array values using the new time-dependent inverse sensitivity (PHOTFLAM) keyword values, align the images in detector coordinates, and combine all images in a given CCD and filter. This process allows for a more accurate measure of the PSF at large radii out to 6 arcsec. We compare the EE values in a 10-pixel radius aperture, EE(10), used for the photometric calibration, and we find that the values for the two CCDs now agree to ~0.1% for each filter, compared to the 2016 solutions which differed by up to 0.5%. At UV wavelengths, the EE(10) is now lower by ~1% for both CCDs, in closer agreement with the 2009 solution. For two ‘red’ filters (F775W and F850LP), we compare the EE curves for a white dwarf and a G-type star but find no significant differences due to the color of the source. The new EE(10) values will be applied in a future update to the UVIS photometric calibration.

We examined the pixel stability and “heat” in the Cycle 28 IR dark files which are used to create the latest IR bad pixel table. Between November 2020 and November 2021 the number of cold and unstable pixels increased by 1.7%, compared to only 0.5% the previous cycle. 91% of cold and unstable pixels were flagged as random telegraph noise and highly variable. The overall median dark current has increased by 6.5% while the error has decreased by 2% in Cycle 28 leading to the increased number of cold and unstable pixels. Trends in dark counts and error levels were checked against a number of telemetry trends but no correlations were found. It is possible that low-level or undetectable changes in detector temperature or voltage could produce the observed 0.08 counts/sec change. We determined that these trends began before the July 14 2021 side-switch of some of the SIC&DH memory modules and electronics and are therefore unrelated to that event. We examined additional potential detector issues and searched for matching and correlating trends but none were found

We use Wide Field Camera 3/UVIS internal flat-fields taken with the calibration subsystem deuterium lamp between the Servicing Mission Orbital Verification (SMOV) in August 2009 and May 2021 to document changes in the ultraviolet filters and to track the performance of the deuterium lamp. The internal calibration subsystem uses a much different f-ratio (∼f/300) than external science images (f/31) and therefore these flat-fields are not used in the calwf3 calibration pipeline to directly correct science exposures. The comparison of the flat-fields taken in December 2020 & May 2021 to those taken during SMOV show a drop of 1-12% in flux (normalized to SMOV) depending on the filter. The drop for the bluest filters (F218W - F280N) is several times higher than for the redder filters implying that the deuterium lamp may be reddening with time. Only ∼2% of the decrease in the ratio images can be accounted for by the UVIS photometric sensitivity losses. One new feature, named the "bowling pin", has been discovered in the flat-fields and the cause remains unknown at this time. The bowling pin is an amorphous region of pixels mostly on the quadrant B side of UVIS1 that shows a decreased signal level compared to the median of the image. The rate of signal decline is increasing over time, and the affected pixel area is also increasing. Additionally, the feature exhibits a color dependence where the affected pixel area and the decrease in signal compared to the median is greatest at shorter wavelengths. The pixels within the bowling pin are ∼1-5% below the median pixel value of the ratio image depending on filter and epoch. All of the previously reported filter and deuterium lamp artifacts remain unchanged: there are no new droplets or dust motes, and the downturn of signal at the outer corner and edge of quadrants B, C, and D is still in agreement with values last recorded by Baggett (2007).

We examine Wide Field Camera 3 UVIS internal flat-field images acquired with the tungsten lamp 3 to check for any changes in the filters over time as well as report on the performance of the lamp. The data for this study were acquired from Cycle 17 through Cycle 28 (May 2009-March 2021). UVIS flat-field images show an overall decrease in countrates from 1-4% over the last 12 years depending on wavelength. Bluer wide-band filters show greater loss in performance compared to redder filters, indicating a possible reddening of the lamp over time. This decrease in brightness is principally associated with the photometric sensitivity losses observed by previous reports as the level of sensitivity loss matches the tungsten flat image ratio levels. Several new artifacts of unknown origin have been noted in the filter ratio images, most notably a dark patchy region in amps A and B for the bluer filters. There is also a very uniformly rectangular dark region (on multiple filters in amp B), whose right edge coincides with one of the lithography mask lines. Both the patch and the rectangular area regions show a decrease in brightness larger than the median value of the filter ratio and typically on the order of 2-5% depending on the artifact. No new dust motes or droplets have been discovered since SMOV in 2009 and the existing particulates have shown that, as noted in prior studies (e.g. Bourque and Baggett 2013), some filter wheels occasionally do not return to precisely the same position and are 1 step off. This offset does not impact external science observations.

The pipeline used to generate and deliver the UVIS bad pixel tables is re-evaluated following the implementation of a 20 electron post-flash on the dark calibration images that led to an anomalous jump to ≈3.0% in the unstable pixel population. We find that the rise in unstable pixels is due to a mis-classification of the hot and cold pixel populations, which led to an inaccurate calculation of the stability threshold. Rather than classifying hot pixels using the ≥13.5 electron hot pixel threshold for a 900s exposure, the updated pipeline now uses a formula to determine the hot pixel threshold that is based off the post-flash amount and dark current on the detector. The new hot pixel threshold is calculated to be 22.7 electrons for the 20 electron post-flashed darks, and results from the updated pipeline show that the anomalous unstable pixel population is brought down from 3.28% (3.07%) to 1.11% (1.16%) for chip 1 (2) for the November 2020 anneal cycle, and from 3.31% (3.08%) to 1.05% (1.08%) for chip 1 (2) for the December 2020 anneal cycle, consistent with previous cycles, before the 20 electron post-flash was implemented. This new method of determining the hot pixel threshold will ensure an accurate measurement of the unstable pixel population, independent of future modifications to post-flash intensity.

The pixel-based CTE correction within the calwf3 pipleline has worked well for many years, but as the CTE-related losses continue to increase in low-to-moderate background images, the algorithm is no longer able to correct faint sources for CTE without inducing extreme amplification of readnoise. Thankfully, there are several ways available to correct for CTE losses in WFC3/UVIS images. The formula-based corrections are still effective for aperture photometry, but as a new photometry routine (hst1pass) is becoming available to measure images and correct photometric and astrometric measurements for CTE losses, we revisit these empirical corrections and put them into a format that is more useful for this application. To this end, we distill the corrections into simple 2-D tables that provides the photometric losses and astrometric shifts as a function of (1) the initial flux and (2) the local sky background. These tables have now been implemented into the soon-to-be-released hst1pass routine.

Accurate and precise measurements of stellar PSFs (Point Spread Functions) have many applications in astronomy. In order to maximize HST's scientific output, we built libraries of observed stellar PSFs from WFPC2, WFC3/UVIS, and WFC3/IR images. We describe the star-image-extraction pipeline, which involves identifying potential stellar sources in calibrated images and measuring against various PSF models. We report that the mean fractional disagreements between model and image pixels were 0.103+/-0.030, 0.090+/-0.076, and 0.046+/-0.024 for WFPC2, WFC3/UVIS, and WFC3/IR, respectively. We summarize the contents of the database by analyzing several distributions and confirming that these distributions are aligned with expectations. All the images of confirmed stellar sources were uploaded to the Mikulski Archive for Space Telescopes (MAST) along with complimentary metadata obtained from engineering and model-PSF fitting. We also discuss how to retrieve appropriate PSFs for specific applications using the archive. Finally, we discuss key astronomical applications of the database, and future work.