Revise snowball correction step to handle partially-saturated grouped data

[stscijgbot-jp]

Issue JP-3835 was created on JIRA by David Law:

As discussed by https://jira.stsci.edu/browse/JP-3805, introduction of flagging for cosmic rays saturating in grouped data meant that the snowball algorithm no longer effectively found all snowballs.  A workaround was implemented for build 11.2 to instead flag cosmic rays saturating in grouped data as DO_NOT_USE instead of SATURATION, but ideally the snowball detection code should be updated to work with saturated grouped data too.

Setting to 'housekeeping' priority, as this does not have scientific impact.

[stscijgbot-jp]

Comment by Christian Hayes on JIRA:

One thing I noted while working on https://jira.stsci.edu/browse/JP-3916 is that the patch fix in https://jira.stsci.edu/browse/JP-3805 also turned off the saturation growth for pixels that saturate in group 2, so there might be a minor science improvement in finding a workaround for the saturation-snowball interaction and turning that saturation growth back on (often the neighbors of these saturating pixels also appear to have heightened rates).

[stscijirahub]

This issue is tracked on JIRA as JP-3835.

[stscijirahub]

Comment by David Law on JIRA:

Adding a quick note that I think I might see some of what's going on.  This is using test data jw01345001001_10201_00001_nrca3, which has 5 groups and 8 frames per group.  Many of the problem cases from build 11.1/11.2 are no longer problems, but one example snowball is at (x,y) = (271, 874).  The central CR occurs in group 2.

The first oddity is that the core of the snowball gave NaN in 11.2, but gives a high value now.  It looks like that's because now we allow use of 1-group data, and that first group is slightly high due to random chance, detector noise, or some similar effect.  The pixel value isn't high because we're seeing the snowball, it's high because it's 1-group data and that's noisy.

The second issue is why the halo of this snowball leaks through when saturated pixels are flagged as SATURATED (i.e., changing the code to do so) as opposed to DNU.  I think the reason why is that:

• In the DNU case, there are enough pixels in group 2 with JUMP set that the snowball code sees them and performs snowball flagging.  In group 3 the core is set as SATURATED, which the snowball code sees as a new set of SATURATED pixels and again performs snowball flagging.
• In the SATURATED case, group 2 triggers both the JUMP and SATURATED criteria, so this is no longer a 'new' SATURATED case in group 3.  As a result less of group 3 is flagged and parts of the halo glow are introduced into the data.

[stscijirahub]

Comment by David Law on JIRA:

Draft PR available at spacetelescope/stcal#576

Tagging Michael Regan FYI in case you have any particular snowball test cases you want to test this out on.