Instrument Science Reports (ISRs)

We present new superdark calibration files for use with IR channel data from Wide Field Camera 3 (WFC3/IR) for HST Cycles 26-30, with all allowed observing modes. These superdarks incorporate five years’ worth of new IR dark exposures, and make use of updated bad pixel tables to more accurately represent the dark current in time-dependent hot pixel populations for data taken during those cycles. The files are available through the Calibration Reference Data System (CRDS) and have been used to reprocess affected datasets for new downloads from the Mikulski Archive for Space Telescopes (MAST).

We align more than 7,400 WFC3/UVIS exposures to the Gaia DR3 catalog to examine the time evolution of the linear terms (shift, rotation, scale and skew) of the geometric distortion solution between 2009 and 2022. We find small linear temporal changes in the scale and skew terms (less than 0.2 pixels in 13 years) which are generally dominated by intrinsic scatter (up to ±0.3 pixels). Concurrently, a larger filter-dependent offset in the scale term is observed, with a maximum difference of 0.3 pixels between F275W and F814W images at all epochs. A small rotation offset to Gaia of 0.003 ± 0.004 degrees is measured from 2009 to mid-2017, after which the offsets are as large as 0.01 degrees, with a large scatter. MAST pipeline processing includes an additional alignment step which corrects UVIS images for any residual linear terms with respect to Gaia DR3 when there are at least 10 matched sources. In addition to any pointing offsets, this step accounts for any evolution in the distortion linear terms described here. For observers requiring high-precision astrometry, we recommend using the tweakreg routine to realign images using a 4-parameter fit (x−shift, y−shift, rotation, and scale) or a 6-parameter fit (x−shift, y−shift, x−rotation, y−rotation, x−scale, and y−scale) depending on the number of matched sources. We provide links to DrizzlePac tutorials for improving both absolute and relative astrometry in WFC3 images.

We present a process for designing and training a convolutional neural network to classify detections from star finding algorithms in the exposure (FLT/FLC) frame for WFC3/UVIS images. This network can be used to filter out common spurious detections (cosmic rays or diffraction spikes), which are often present in imaging data where the point spread function is under or critically sampled (FWHM = 2 pix). The neural network achieved high accuracy for correctly identifying stars (96%). Falsely detected sources in catalogs (a common side effect of DAOFind-like algorithms) can cause incorrect matches and resulting astrometric transforms. Eliminating false detections can provide significant improvements to image alignment workflows and clean photometric catalogs. We also present caveats of this method, as well as general considerations for building neural networks for use in astronomical data analyses.

We present new time-dependent inverse sensitivities for the WFC3/IR channel. These were calculated using the sensitivity change slopes measured by Marinelli et al. (2024) and photometry of five CALSPEC standards (the white dwarfs GRW+70 5824, GD 153, GD 71, G191B2B, and the G-type star P330E) collected from 2009 to 2023. The new inverse sensitivities account for losses of 1-2% over 15 years, depending on wavelength, and provide an internal photometric precision better than 0.5% for all wide–, medium–, and narrow-band filters. An updated version of calwf3 (v3.7.2) has been developed for use with a new time-dependent image photometry table (IMPHTTAB) and will be used to update the image header photometric keywords following MAST reprocessing, expected in late-2024. Alternatively, the new inverse sensitivities may be computed by the user for a specific observation date by running stsynphot.

The pixels on WFC3/UVIS can experience low quantum efficiency (QE), in which a pixel’s sensitivity decreases by more than a percent typically for a few days. Annealing is a strategy for recovering pixels to nominal sensitivity. To characterize these low QE pixels and their relationship with annealing, we obtained a decade’s worth of internal flat-fields from the deuterium lamp for UV filters (F225W and F336W), and from the tungsten#3 lamp forVIS filters (F438W and F814W). We also corrected for UV time dependence and disregarded pixels affected by artifacts, such as droplets and dust motes. Our results generally agreed with past reports: low QE pixels were time- and wavelength-dependent.Depending on the bandpass, 0.1-1% of pixels were actively low QE at any given time, but 23-62% of UVIS’s pixels experienced low QE at least once. The low QE population grew within anneal periods, but anneals recovered greater than 90% of low QE pixels to nominal sensitivity. Low QE pixels were typically isolated spatially and temporally, but can experience rare effects such as clumping, flickering, and losing sensitivity permanently. Low QE populations were also unique sets of pixels at each epoch. Contrasting to previous studies, we found that F438W lost the most QE on average than bluer or redder filters, suggesting a more complex wavelength dependence than a simple linear relationship. We recommend dithering to mitigate low QE pixels. In addition, we used Amazon Web Services and ChatGPT in this work so we provide context and insights on their viability in astronomical operations.

Here we characterize the time-dependent sensitivity evolution of the Hubble Space Telescope’s Wide Field Camera 3 (WFC3) UVIS G280 grism, which provides spectroscopy between 1900-8000 Å with a resolving power of R∼70 at 3000 Å. Using archival G280 observations of the flux standard white dwarf GRW+70D5824, we find wavelength-dependent sensitivity losses between 2011-2024 ranging from −0.188±0.026% to −0.347±0.043% per year across eight synthetic bandpasses spanning the full wavelength range of the grism. These losses are greater than those measured from photometric monitoring campaigns of this target using staring mode and spatial scanning observations. The differences in the sensitivity evolution between the UVIS photometry and G280 grism data – which span comparable wavelength ranges – may be driven by construction differences (e.g., differences in the coatings or substrates) between these optical elements.

In this report, we analyze all WFC3/IR internal flat field calibration programs through 2024, updating trends in count-rate and sensitivity effects. We update the analysis of Ryan (2019) to include 13 cycles of WFC3/IR internal flat field images, up to Cycle 31. We follow a similar post-analysis masking technique to the earlier study, filtering out pixels affected by persistence up to 6 days prior. We confirm that the mode count-rate of each filter in WFC3/IR is decreasing with time at about 0.25 - 0.40% yr⁻¹, 0.31 - 0.33% yr⁻¹ averaged over all filters. However, the rate of decrease has become relatively less stable in recent cycles; we note a small (∼ 0.7% over expected) increase in count-rate in Cycle 31 in all filters. More data will be needed to confirm this increase. As in the previous study, the count-rate decrease is wavelength dependent, although the additional data has considerably tightened the trend. The previous study had noted that F160W and F153M were nearly unaffected by the count-rate decrease. In contrast, the additional data in this report restores the count-rate decrease in F160W and F153M back in line with other filters. Despite this correction to the reddest wavelength filters, the yearly decrease is still larger at bluer wavelengths, and the count-rate increase in Cycle 31 is greater at redder wavelengths. Both trends support the proposed reddening tungsten lamp from earlier studies, although a similar (in time and magnitude) increase in photometric sensitivity was also noted in globular cluster observations from other studies. Ultimately, the changes we measure in the internal flat fields are a combination of the reddening lamp and the changing overall sensitivity in the IR detector. Shortly after these observations, HST entered reduced gyro mode, and a future report will determine which trends are affected.

We continue previous studies of Dragon’s Breath and Scattered Light anomalies on the WFC3/UVIS detector with an additional 764 Dragon’s Breath flagged images (on top of 1,338 images in the original study) from September 2016 to February 2024. The percent of images affected by Dragon’s Breath (between 6% to 10%) in the past and current dataset is consistent within the margin of error for all tested filters (F606W, F814W, F200LP, F350LP, and F600LP) except F200LP, which has a significantly lower occurrence rate in the new dataset (from 13% down to 4%). However, this drop in occurrence rate is most likely due to small statistics. Stars causing Dragon’s Breath continue to peak in appearance within 490 pixels (∼20 arcseconds) off the edge of the detector. Dragon’s Breath seems to appear relatively uniform around the detector though somewhat less often around the Amp A and Amp D corners. Dragon’s Breath stars peak within 0.5 magnitudes of 11.9 and 14.6 V-band magnitude across all filters. The Dragon’s Breath tools on the WFC3 anomaly page, including the interactive bokeh plot and the searchable table are updated with the new dataset up to February 2024. In addition, a new XML overlay plug-in for APT has been published to assist users in identifying stars that may cause Dragon’s Breath when preparing their proposals.

We describe some details related to changes in image scale induced by velocity aberration of starlight. We investigated whether an update was warranted to the HST keyword VAFACTOR, which quantifies the dimensionless factor by which the image scale is changed by velocity aberration at the mid-point in time of an exposure. Nominally, the factor differs from unity by ±1E-4 or less. We quantify each of the following effects on VAFACTOR: a) using the classical formula instead of the relativistic one, b) evaluating it at an inappropriate time, e.g. at the start of an exposure, and c) evaluating it at mid-exposure instead of averaging it over the entire exposure. Respectively, the three effects can produce errors in VAFACTOR as large as a) 1.0E-8, b) 0.8E-6 T, and c) 4.5E-9 T², with the exposure time T expressed in minutes. Whereas the first effect is negligible and the second effect is only hypothetical, the third effect has the potential to be marginally significant in rare circumstances, but we expect will never matter in practice. Based on this analysis, no updates are indicated to the formulas used for FITS keywords VAFACTOR (HST) or VA SCALE (JWST).

Most discussion of charge-transfer-efficiency (CTE) losses involves the parallel transfer of charge in the y-direction down the chip columns (y-CTE). Serial charge-transfer efficiency (x-CTE) refers to the horizontal transfer of charge. Serial CTE losses were first assessed in WFC3/UVIS in 2014 (WFC3/ISR 2014-02), where it was found that bright stars were shifted by ~0.0015 pixel away from the readout amplifier. Now that the WFC3/UVIS CCD has spent more than three times as long in space, imperfect serial CTE should have three times the impact on the images, so we revisit the effect, characterizing serial CTE in a similar manner to our models for parallel-CTE by using a combination of warm pixels, cosmic rays, saturated stars, and overscan pixels. Overall, serial CTE has a much smaller impact on images than does parallel CTE. As of 2023, parallel CTE can have a ~5% impact on bright sources and a >50% impact on faint sources, and serial CTE can have a ~1% impact on bright sources and a ~3% impact on fainter sources. The first few pixels in the serial CTE trails are much sharper than the parallel trails, but there is a very faint component to the serial trails that extends much farther than the parallel trails — even wrapping around to the next row. We develop a pixel-based model for the trapping and release of charge in the serial register and release a stand-alone beta-version of this pixel-based serial-CTE correction. Most images are not significantly impacted by the x-CTE effect, however HST users that require high-precision astrometry could benefit from this correction — at the very least, so that they can quantify its impact on their science.

The WFC3/IR detector has experienced a cumulative photometric sensitivity loss of ~1 - 2 % since installation, indicating that time-dependent corrections are necessary for highest accuracy photometry. We evaluated the changing photometric sensitivity using staring mode observations of globular clusters, scanning mode photometry of an open cluster, and grism observations of four CALSPEC standard white dwarf stars. Staring mode observations of CALSPEC standards were used for testing and validation. Sensitivity change appears to be wavelength-dependent, with the greatest losses in the bluest filters. A new IMPHTTAB reference file, with updated inverse sensitivities, will be delivered in 2024. In the interim, we provide factors users can apply to manually correct their photometry.

We present phase retrieval measurements of the focus of the Hubble Space Telescope (HST) found using the WFC3/UVIS and ACS/WFC detectors from February 2019 to December 2023. From our analysis we predict that by the end of 2024 WFC3/UVIS will be at -0.66 micron despace and ACS/WFC will be at -1.48 micron despace.

This report examines Charge Transfer Efficiency (CTE) flux losses in the Wide Field Camera 3 UVIS detector aboard the Hubble Space Telescope. Spanning approximately 14 years of observations from October 2009 to February 2024, the study analyzes CTE flux loss trends across various total background levels and source fluxes. In addition to analyzing the present state of CTE flux losses, we provide updated coefficients for the empirical model for point source photometry corrections in both non-CTE-corrected (FLT) and CTE-corrected (FLC) data. Between 2009 and 2023, the rate of CTE flux loss for a 500-2000 e− source, farthest from the readout, in FLT data with a 1-3 e−/pix background, is measured to be ∼0.05 Δmag/2051 pix/year. The recommended minimum total background level to mitigate CTE losses remains at 20-25 e−/pix. At that level, we find that 500-2000 e− sources, farthest from the readout, in 2024 FLT data can suffer ∼23% flux loss/2051 pix. The FLC data provide some relief, but we measure a ∼12% flux loss/2051 pix in 2024. There continues to be a slight over-correction in some FLC results that contain backgrounds above 40 e−/pix. In 2024, 8000-20000 e− sources farthest from the readout in a 40, 60, or 90 e−/pix background are over-corrected by ∼1, 2, and 3%, respectively.

The Wide Field Camera 3 (WFC3) onboard the Hubble Space Telescope (HST) has captured over 310,000 images in its near 15-year lifetime. Some of these images are affected by guide star failures, which can cause a smearing of the sources in the image. Although the images are manually flagged by WFC3 team members for such anomalies, machine learning is more practical for observatories that will be far more data rich, and where manual flagging will be inefficient or even impossible. In order to remedy this problem, we trained a convolutional neural network (CNN) to identify WFC3/UVIS images affected by guide star failures. The CNN's training and validation data were taken from May 2009 to May 2022. We developed a data processing pipeline to log-scale, down-sample, and normalize the images. Our best model achieved true negative and true positive rates of 90% and 91% on our validation data. We investigate the model's misclassifications, deployment tests, and rotational dependency. In addition, we present shortcomings from other trained models and ideas for future work. Our code and model parameters can be found on Deepwfc3's GitHub.

Our study focuses on the enhancement and validation of IR bad pixel tables (IR BPIXTAB) for Cycles 29 and 30, along with the revision and distribution of two existing IR BPIXTABs to rectify anomalies observed in 2021 and 2022. We address the issue of an unusual spike in cold and unstable pixels detected during Cycle 28, attributing it to algorithmic discrepancies within the pipeline utilized for BPIXTAB generation. Through meticulous analysis, we identify that the variance in input darks, sourced from three distinct BPIXTAB reference files, significantly impacts the prevalence of flagged cold and unstable pixels. By generating three separate IR BPIXTABs for Cycle 28 with consistent reference files, we observe a substantial reduction in the flagged pixel percentage (from 2.85% to less than 1%), affirming the presence of an algorithmic error rather than an intrinsic detector issue. Employing the most recent accurate IR BPIXTAB, labeled 53514239i_bpix, we successfully mitigate the abnormal pixel spikes observed in Cycle 28, ensuring consistency across subsequent cycles. Furthermore, we provide an updated Cycle 28 IR BPIXTAB, utilizing input darks referencing 53514239i_bpix, to reflect the corrected cold and unstable pixel counts.