Data artifacts known as snowballs (in the near-IR) and showers (in the mid-IR) are caused by large cosmic ray impacts that affect hundreds of pixels (M. Regan 2023). 

The vast majority of cosmic ray impacts directly affect only one or 2 pixels. However, snowballs and showers affecting hundreds of pixels are often seen in the data as well. In the near-IR detectors snowballs tend to be circular, while in the mid-IR detectors showers often have elongated and irregular shapes.

A very rough estimate of the rate is around one snowball every 20 s in each 2K × 2K near-infrared detector, with approximately a factor of 2 lower rate of occurrence for shower events in the mid-IR. This frequency is highly variable however, with some observations affected by large numbers of such artifacts and other observations nearly completely unaffected. While similar events were seen in the near-IR detectors on the ground, the typical frequency was an order of magnitude lower and the amount of charge deposited was also significantly smaller.

Both snowball and shower features are difficult to detect in the pipeline, as the deposited charge at large distances from the primary impact site can be small and released gradually over time rather than in a single event (in some cases even persisting through a detector reset between exposures). These features are most significant for deep observations, and can limit the ability of such observations to reach the detector noise floor.

Starting with jwst pipeline build 9.0 (Fall 2022) the jump step in the pipeline has been capable of detecting and flagging pixels affected by both showers and snowballs, and the algorithms for doing so have been continuously improved since.  As of build 10.2 (Spring 2024) snowballs in the near-IR detectors can be almost entirely removed, and the pipeline does so by default for NIRCam, NIRSpec, and some NIRISS observing modes.  Showers in the mid-IR can be improved (although not removed entirely), which is done automatically for some MIRI imaging modes.  Users can toggle this step for themselves in offline reprocessing using the find_showers (mid-IR) and expand_large_events (near-IR) parameters respectively (see the ReadTheDocs Jump Detection page.)

---

Snowballs

In the near-IR, snowball artifacts are produced by extremely energetic cosmic rays that saturate the detector and affect the slopes of neighboring pixels. Figure 1 shows a close-up of a slope image (generated in stage 1 detector processing) centered on the location of a very large snowball. This slope image is from a long NIRSpec dark exposure and is very sensitive to any perturbation in the calculated slope. Thus, this is a worst case example; images with significant background rates will see much smaller effects.

Since dark exposures accumulate counts in a highly linear fashion, they provide a sensitive probe of deviations caused by the event. Four regions can be identified. 

• The central region inside the white circle indicates pixels that saturated when the particle hits the detector (allowing ramp fitting to fit only groups observed prior to the hit). 
  
• The white ring of pixels just outside of the saturated pixels represents pixels that saw a discontinuity in the count rate when the event occurred but did not saturate. The discontinuity was flagged as a "jump" by the pipeline and was not included in the slope calculation. They show the excess slope due to charge spilling from the saturated region in subsequent groups of the integration.
• The large darker region outside the white ring is caused by pixels that were flagged as a jump when the snowball occurred but then had below average count rates because photoelectrons were temporarily trapped by small defects in the detector pixels. 
• Finally, the outer white halo represent pixels that saw an increase in the accumulated charge when the snowball event occurred but were uncorrected because the increase was below the pipeline's detection threshold.  

Although the halo regions are clearly indicated in dark exposures, they can sometimes be difficult to see in images with significant astrophysical or other backgrounds.

As indicated by Figure 2, the majority of snowballs are relatively small, affecting up to about 100 pixels. There is a long tail to the distribution though, with some snowballs impacting nearly a thousand or more pixels.

The algorithms to detect and remove these snowball artifacts from the data have improved throughout the mission. As illustrated in Figure 3, the typical science data products provided by build 10.0 of the jwst pipeline contained multiple artifacts as the algorithm was not robust enough to be turned on in default processing. Build 10.1 improved the ability of the algorithm to flag faint outer regions of snowballs substantially, although leaving persistence from the bright cores. Build 10.2 in turn addressed both these residual bright cores and multiple other small improvements, allowing for the step to be run by default on most NIR observing modes.

Ongoing work will address further remaining issues, including the relatively uncommon case of non-circular snowballs likely arising from cosmic rays with more elongated paths through the detector.

---

Showers

In the mid-IR, shower artifacts are thought to result from large clusters of energetic particles produced by cosmic rays. Unlike traditional cosmic rays, which affect a small number of pixels and produce a nearly delta function jump in the ramp for a given pixel, these cosmic ray showers can affect hundred of pixels (see Figure 1), release charge slowly over a period of many seconds, and can produce latents visible in the next integration (and in some cases are produced by initial events that occurred while the detector was in idle mode before a given exposure began).  Like snowballs in the near-IR, showers can be strongly time variable and likely depend on the solar weather.  Unlike snowballs however, shower artifacts typically have irregular, extended shapes and are significantly more challenging to detect and flag in the jwst pipeline.

Shower artifacts in MIRI MRS data

Cosmic ray shower artifacts typically have count rates in the range of 0.03 to 0.5 DN/s/pixel.  They are thus most noticeable in MIRI MRS observations of faint targets (particularly in channels 1 and 2), where they can produce a mottled pattern across the detector that changes from exposure to exposure.

The jwst calibration pipeline is able to detect and correct some of these shower artifacts in the data by flagging the affected groups in the ramp using the find_showers parameter in the jump step of the calwebb_detector1 pipeline.  However, this still leaves significant structure remaining that can limit the performance of the MRS in the faint-source regime.  As of build 11.3 an additional optional correction is available by using the clean_showers flag in the straylight step of the calwebb_spec2 pipeline.  As illustrated in Figures 4 and 5, this optional correction can significantly improve performance for faint sources.

Shower artifacts in MIRI Imaging data

As illustrated in Figure 6, cosmic ray showers can also be observing in MIRI imaging data.  The most obvious such cases occur when using the F560W filter as the zodiacal background is low at these wavelengths and detector effects more noticeable.  At longer wavelengths where the background is higher the low count rate of showers tends to be less noticeable in comparison.  Generally, such artifacts can be reasonably well removed through a combination of the find_showers parameter in the jump step along with the outlier_detection step when combining multiple dithered exposures.

---

References

Regan, M. 2023, JWST-STScI-008545
Detection and Flagging of Showers and Snowballs in JWST

Dicken, D. et al. 2024, A&A, 689, 5
JWST MIRI flight performance: Imaging

Spilker, J. et al. 2023, Nature, 618, 708
Spatial variations in aromatic hydrocarbon emission in a dust-rich galaxy

---

---