7.5 Blobs

Small features called “blobs” are present in all WFC3/IR images and may be noticeable in observations with high background or in observations of large, extended objects. These artifacts affect ~2% of IR detector pixels and are caused by de-focused particulates on the surface of the mirror mounted on the Channel Select Mechanism (CSM), which is used to switch between UVIS and IR channels. Blobs are typically 10-20 pixels in diameter and can reduce the incoming light by up to ~30% in their central pixel, although the effect is typically much less.  Since the CSM mirror is moved out of the beam in order for light to enter the UVIS channel, blobs do not appear in UVIS images (although UVIS data contain a different set of artifacts referred to as “droplets” (cf. Section 5.4.1)).

New delta-flat reference files were delivered in October 2020 to correct for blobs during calwf3 processing for six IR filters: F098M, F105W, F110W, F125W, F140W, F160W (cf. Section 7.5.3 and Section 7.8.4). These flats correct for the majority of blobs appearing in IR images through July 2018 (see Figure 7.10 and WFC3 ISR 2021-10 and include 148 of 153 total known blobs cataloged in WFC3 ISR 2018-06. 

7.5.1 Blob catalogs

Regions of low-sensitivity were first noticed in IR flat fields acquired during ground testing (WFC3 ISR 2008-28) and were described as 'small particulate features with reduced QE' (quantum efficiency). The authors recommended that on-orbit images be used to properly flat field these features, which were expected to shift position following the in-flight IR channel optical alignment.

Blobs were observed in inflight images shortly after WFC3 was installed on HST, and WFC3 ISR 2010-06 provided an initial catalog of position and radius for 19 blobs appearing through March 2010. This report also describes a set of calibration test data which confirmed that blobs are located on the CSM and not on the IR detector itself. WFC3 ISR 2012-15 expanded the catalog to 40 total blobs appearing through July 2012 and included a new estimated ‘date of appearance’ and blob ‘color’ which reported the strength of the blob in different filters. The blobs were found to be more opaque to bluer light and their effect is therefore slightly stronger in the F105W and F125W filters compared to the F160W filter. 

Starting in mid-2010, calibration observations of the dark Earth were used to identify many smaller and weaker blobs which were not apparent in science images. By using airglow as a uniform sky background illumination in the F153M filter, the catalog was expanded to 127 blobs through March 2014. WFC3 ISR 2014-21 describes the procedure for detecting blobs and for measuring their 'strength' based on the integrated sensitivity loss over their total area. Identification numbers were assigned to each blob in the three strength categories. Of these, only the 46 strongest blobs (losses exceeding 2.6%) were included in a new set of time-dependent bad pixel tables, with a separate table for each unique appearance date. The remaining 81 weaker blobs were cataloged but were not assigned any bad pixel flags.

The blobs have been increasing in number monotonically in time; no blob, once it appears, has disappeared. The rate of appearance was higher in the first year after launch (2009) than it was later. To monitor the appearance of new blobs, the schedulers insert dark Earth flats into appropriate gaps in the observing schedule, providing typically ~2-3 observations per week (e.g. WFC3 program CAL-17371). The catalog was updated again in 2018 (WFC3 ISR 2018-06) to include four additional years of calibration data (WFC3 ISR 2018-06). For convenience, blob IDs were renumbered according to their estimated ‘Appearance Window’ rather than by their strength, so that new blobs may simply be assigned the next consecutive number. The catalog in this report (WFC3 ISR 2018-06) is continually updated when new blobs appear. As of May 2024, there are 153 known blobs, with 58 classified as strong (Figure 7.4).

Since all new blobs are manually flagged by a WFC3 team member, we aimed to automate this process using machine learning. A convolutional neural network (CNN) was built, trained, and evaluated to classify if a new blob was present in 256x256 cutouts of IR images (WFC3 ISR 2021-08). This first step in automatic anomaly detection explored how machine learning can be used in astronomical operations, especially in the dawn of astronomy’s era of big data. The model’s accuracy was 91%, and is available for use on the DeepWFC3 GitHub repository. 

7.5.2 Blob Map

The CSM mirror is near to and slightly tilted with respect to the telescope's focal plane. As a result, blobs’ radii systematically increase from the detector’s upper right corner (where they are nearly in focus, with radii less than ~4 pixels) toward the lower left corner (where they have radii of ~13 pixels). Figure 7.4 provides a map of blobs on the IR detector, color coded by their strength, where green and red circles highlight the strongest blobs flagged in the bad pixel tables.

Possible overlaps between known blobs and the IR grism aperture reference positions and the IR dither patterns have been investigated (WFC3 ISR 2017-16). Apertures are designed to place the science target on a cosmetically clean area of the IR detector. Similarly, dither patterns are designed to mitigate cosmetic defects by rarely (ideally never) placing such targets on known defects.  As reported in (WFC3 ISR 2017-16), only two potential overlaps were found but no changes were made to the defined apertures or dither patterns because 1) one of the overlaps occurs with a dither/aperture combination that is most commonly used for wide-field surveys/mosaics and 2) the other overlap is ~9 pixels from a blob that has a radius of ~10 pixels i.e. an already conservative distance. Since blobs accumulate over time, the originally defined apertures and dither patterns are periodically checked to verify that they still accomplish their goals for science observation.

7.5.3 Blob Correction Strategies

Appropriate dithering during Phase II planning (e.g. PATTERN=‘WFC3-IR-DITHER-BLOB’ or a custom ~5” dither) can permit the cleaning of blobs from combined images during advanced reprocessing, as described in WFC3 ISR 2010-09. Nominal drizzled images (*drz.fits) produced by MAST ignore the blobs as flagged in the DQ array of calibrated (*flt.fits) images, assuming that not every program utilizes a 'blob dither'. Even with effective dithers (+/- 15 pixels), the images cannot maintain pixel-phase coherence. Multiple sets of four-point dithers are necessary to compensate for both blobs and pixel-phase coverage (WFC3 2023-05).

An alternative method of correcting for blobs is to apply a delta-flat correction image, the first of which was derived from dark Earth observations (WFC3 ISR 2014-21). This correction was initially recommended for cosmetic purposes only; the authors suggested that while it may improve the photometry of extended sources, it was not expected to work well for stellar photometry due to the small optical cross section of the blobs. However a follow-up study that applied a delta-flat to dithered exposures in a crowded region of 47 Tucanae determined that applying a blob flat field is effective for improving stellar photometry in blob-affected areas (WFC3 ISR 2015-06). 

In October 2020, a new set of delta-flat (D-flat) reference files were delivered to CRDS to correct for blobs during pipeline processing for six IR imaging filters: F098M, F105W, F110W, F125W, F140W, F160W (WFC3 ISR 2021-10). These flats were computed by stacking observations of sparse astronomical targets after the appearance of each blob to produce a deep image of the sky background (see Section 7.8.4). While the bad pixel tables continue to carry DQ flags for strong blobs (green and red in Figure 7.4), the delta-flats now provide DQ flags for all cataloged blobs, regardless of their strength. This increases the total number of flagged pixels (DQ= 512, 'bad in flat') from 1.1% to 2.3% in calibrated IR (non-grism) images. The D-flats correct for 148 of the 153 known blobs at 49 unique “appearance dates.” Additional D-flats will be delivered for the most recent blobs once more inflight data is available.

The new D-flats (*_dfl.fits) are intended to be used together with an updated set of P-flats, which improve the 'pixel-to-pixel' sensitivity calibration with wavelength for all imaging filters (see Section 7.8.3). As part of the flat field calibration step, calwf3 uses the product of the P-flat and D-flat reference files to correct for spatial variations in the detector sensitivity with time. The D-flats represent an average of the CSM angle positions sampled with in-flight images, and any given image may contain blobs at an off-nominal position, producing the characteristic dark and light residuals in Figure 7.5. In this case, users may elect to ignore both the delta-flat correction and the additional DQ flags for weaker blobs by setting the DFLTFILE keyword to 'N/A' in the header of the raw image and then reprocessing with calwf3. Details on manual reprocessing of WFC3 data are provided in Section 3.5.

After comparing a large set of dark Earth flats acquired over time, slight offsets in the blob positions were discovered and attributed to changes in the CSM position when switching from UVIS to IR. To explore this effect, a set of seven delta-flats were computed by stacking images at slightly different positions, in increments of 0.004 degrees and spanning a total range of 0.024 degrees of rotation (WFC3 ISR 2021-10). This total range corresponds to a diagonal shift of ~1 pixel for blobs near the center of the detector. The left panel of Figure 7.5 shows a small region of the 2014 delta-flat, highlighting two blobs with radii ~10 pixels and a total separation of ~100 pixels. To illustrate the effect of flat-fielding an image acquired at an off-nominal CSM position with a delta-flat from the nominal position, the center and right panels of Figure 7.5 show the ratio of two delta-flats with CSM angles differing by 0.012 and 0.024 degrees, respectively. Due to the unpredictability of the <1 pixel blob shifts, it is not possible apply shift-dependent blob delta-flats in the automated calibration pipeline; these would need to be applied manually by the user during recalibration. Alternatively, observers concerned about the blobs can consider expanding the outer rim of the existing flagged blobs in the DQF; while this will unavoidably flag some good pixels, it will cover the extent of shifts seen to date.