Details
-
Type:
Story
-
Status: Done
-
Resolution: Done
-
Fix Version/s: None
-
Component/s: skymap
-
Labels:
-
Story Points:6
-
Epic Link:
-
Team:Data Release Production
-
Urgent?:No
Description
As part of metadetect (and other uses!) we will be doing coadds in small edge-free coadd "cells". This ticket adds cells to patches in skymaps as a configuration option.
When this configuration is chosen, the patches are built up in units of cells, and the tracts are built as units of patches, so all patches and tracts contain an integer number of cells (default is 20 cells per patch in the inner region). All patches have an extra ring of cells around the inner region which sets the overlap boundary. This is very helpful to ensure that neighboring patches will have a subset of pixels that are exactly identical (unlike the current "legacy" tract building method).
Attachments
Issue Links
- relates to
-
- Done
Activity
Following up on some discussion, I now see that the DC2 skymap has the same settings as the HSC skymap except for the pixel scale. So I believe the configs were just copied over, with the new pixel scale, and the patches just came out of that. This is definitely something we can revisit for the real LSST survey.
After discussion with Arun Kannawadi we realized that the previous idea that the cells border extends beyond the patch outer boundary doesn't work: if we want to be able to use patch warps as inputs then these need to cover all the area that is covered by the patch outer bounding box. These are what the bounds look like now:
Also, after discussions with Arun Kannawadi and Jim Bosch I have added a bunch of features to the tracts, patches, and cells. In particular, I added properties (so that you can do .innerBBox instead of .getInnerBBox()); I added a reference to the tract wcs to the patches and cells (so you can get polygons and do conversions without the tractInfo!); and I added the sequential patch and cell index to PatchInfo and CellInfo (so you can just access these with .sequentialIndex rather than computing them using the parent).
I think this is ready for review, and I'd appreciate it if both of you can take a look. I'm not sure how to split it up, though, just that I'd like two sets of interested eyes on the changes.
I've left some minor comments on the PR and marked it as Approved; you've already addressed my more substantive ones. Removing myself from the Jira reviewers list.
I've left a number of minor comments on the PR, but overall looks great! Marking the review complete for now, but I plan have one more quick look after the changes are incorporated.
All comments have been addressed, and all the ci's passed: https://ci.lsst.codes/blue/organizations/jenkins/stack-os-matrix/detail/stack-os-matrix/35351/pipeline
Here is some code which will plot a tract + patches and patch + cells: plot_cells_example.py
First, for DC2. The old version is called "legacy", the new version is "cells". The blue dots show the "inner" region of the tract. Looking at the legacy and cell version, the skymap tiling + pixel size just happens to give a lot more "extra" overlap with the cells tiling:
And zooming in on the corner you can see that the legacy patches don't have an outer boundary, but the cell patches do:
Looking at the cells in a patch of course we don't have a legacy version. There is one cell of buffer all the way around:
Now for HSC, with the HSC pixel scale the cell version gives less wasted overlap. Huh!
Zooming in:
And the cells in a patch look very similar to DC2, as they should!