What exactly does helioviewer.org assume about WCS? #494

Activity

1. dgarciabriseno commented on Jun 3, 2026

   

   dgarciabriseno

   on Jun 3, 2026 via email

   Contributor

   More actions

   Hello, sorry I didn't have time to review this today, but I will try to have a better response on Tuesday next week. Off the top of my head, it's likely that the 3D view understands WCS better than the 2D view. I am pretty sure you're right in observing that 2D assumes the same CDELT is the same for both axes, which is the wrong assumption. The files you're likley interested in seeing is JP2ImageXMLBox.php, that's the file that processes WCS info, and it does look like it ignores CDELT2. I suppose we haven't noticed since most data we've added in ourselves has been square.

   …

   On June 3, 2026 7:56:23 AM UTC, salim HACHEMAOUI ***@***.***> wrote: SalimHachemaoui created an issue (Helioviewer-Project/api#494) We are investigating a SPICE JP2 rendering issue in Helioviewer Web. The tested SPICE JP2 is a valid 2D raster: ```text NAXIS = 2 NAXIS1 = 192 NAXIS2 = 612 ```` However, its WCS metadata is not a simple square-pixel, axis-aligned case: ```text CUNIT1 = deg CUNIT2 = deg CDELT1 = 0.0011111111111111 deg ~= 4.0 arcsec/pixel CDELT2 = 0.000305 deg ~= 1.098 arcsec/pixel ``` It also contains a non-trivial PC matrix: ```text PC1_1 = -0.61227620057 PC1_2 = -0.217031764022 PC2_1 = 2.88030582543 PC2_2 = -0.61227620057 ``` This raises a general question: **what WCS assumptions does helioviewer.org make in 2D and 3D rendering?** ## Observed behavior In Helioviewer Web: * SPICE can appear mis-scaled or vertically stretched in 2D. * In 3D, SPICE does not appear correctly overlaid on EUI. * The behavior differs from JHelioviewer, where SPICE appears to be rendered more consistently. ## Main questions 1. Does Helioviewer Web assume `CDELT1 == CDELT2`? 2. Does Helioviewer Web assume that the image X axis, associated with `CDELT1`, is parallel to `solar_x`? 3. Is `CROTA2` used in 2D or 3D rendering? 4. Is the `PC` matrix used in 2D or 3D rendering? 5. Are `CTYPE1/CTYPE2` projections, such as `HPLN-TAN/HPLT-TAN`, supported in Web rendering? 6. Do 2D and 3D use the same WCS assumptions? 7. Is general WCS support expected in helioviewer.org, or only simplified image positioning? ## 2D / 3D context From code inspection, the 2D Web path appears to use the tiled rendering pipeline: ```text getClosestImage -> getTile -> PNG tiles positioned in CSS ``` The 3D Web path appears to use a separate pipeline: ```text resources/js/3d -> @helioviewer/sun -> getClosestImage/getJP2Header/downloadImage -> Three.js texture ``` <img width="1364" height="730" alt="Image" src="https://github.com/user-attachments/assets/7417a145-1a9b-4beb-809e-bad6b453261f" /> -- Reply to this email directly or view it on GitHub: #494 You are receiving this because you are subscribed to this thread. Message ID: ***@***.***>
2. SalimHachemaoui commented on Jun 4, 2026

   

   SalimHachemaoui

   on Jun 4, 2026

   ContributorAuthor

   More actions

   Thanks Daniel, that confirms what we observed.
   For SPICE, the tested JP2 has CDELT1 ~= 4.0 arcsec/pixel and CDELT2 ~= 1.098 arcsec/pixel, so if JP2ImageXMLBox.php ignores CDELT2 and uses a scalar scale, that likely explains the 2D vertical stretching.
   I’ll focus on JP2ImageXMLBox.php for the 2D case and keep the 3D path separate, since it seems to handle WCS differently.

   If you think this is the right direction, I can create a dedicated branch to handle these WCS edge cases, especially CDELT1 != CDELT2 and possibly CROTA/PC later, while keeping the current behavior unchanged for standard square-pixel datasets.
3. dgarciabriseno commented on Jun 4, 2026

   

   dgarciabriseno

   on Jun 4, 2026 via email

   Contributor

   More actions

   That would be greatly appreciated! Yes, the 2D and 3D cases are separate since 3D processes WCS on its own, it is passed to the GPU via a shader to scale/orient the image. Please note HelioviewerCompositeImage.php will also need to be updated to handle CDELT2, this handles making images/frames with the screenshot and movie feature.

   …

   On June 4, 2026 7:50:14 AM UTC, salim HACHEMAOUI ***@***.***> wrote: SalimHachemaoui left a comment (Helioviewer-Project/api#494) Thanks Daniel, that confirms what we observed. For SPICE, the tested JP2 has CDELT1 ~= 4.0 arcsec/pixel and CDELT2 ~= 1.098 arcsec/pixel, so if JP2ImageXMLBox.php ignores CDELT2 and uses a scalar scale, that likely explains the 2D vertical stretching. I’ll focus on JP2ImageXMLBox.php for the 2D case and keep the 3D path separate, since it seems to handle WCS differently. If you think this is the right direction, I can create a dedicated branch to handle these WCS edge cases, especially CDELT1 != CDELT2 and possibly CROTA/PC later, while keeping the current behavior unchanged for standard square-pixel datasets. -- Reply to this email directly or view it on GitHub: #494 (comment) You are receiving this because you commented. Message ID: ***@***.***>
4. SalimHachemaoui commented on Jun 4, 2026

   

   SalimHachemaoui

   on Jun 4, 2026

   Last edited by SalimHachemaoui

   ContributorAuthor

   More actions

   Thanks , that helps a lot.
   I’ll keep the 2D and 3D cases separate. For now I’ll focus on the 2D path and the scalar CDELT assumption.
   I’ll include JP2ImageXMLBox.php for reading CDELT1/CDELT2 and CUNIT1/CUNIT2, and I’ll also check HelioviewerCompositeImage.php so screenshots and movie frames handle CDELT2 consistently.
   The goal will be to keep the current behavior unchanged for standard square-pixel datasets such as AIA/HMI/EUI, and only enable the new logic when the image has anisotropic WCS scale, for example CDELT1 != CDELT2.
   I’ll prepare this in a dedicated branch and share the approach before going too far with the implementation.
5. ebuchlin commented on Jun 15, 2026

   

   ebuchlin

   on Jun 15, 2026

   Last edited by ebuchlin

   More actions

   Thank you very much for this discussion. Thinking a bit more about it with the IAS team this morning:

   • We have the option to bypass the issue by reprojecting the SPICE file to a simple WCS with CDELT2=CDELT1 and PCi_j = Id before creating the JP2. This can be a quick fix for SPICE, but it is not very satisfactory on the long term
   • The reprojection could be done server-side, but this adds some complexity and server load.
   • The web app could switch to the 3D mode when the metadata include CDELT1 ≠ CDELT2 or PCi_j ≠ Id
   • The 2D mode could be deprecated (maybe there are still issues with the 3D mode that prevent this?)

   What do you think of these options?
6. dgarciabriseno commented on Jul 9, 2026

   

   dgarciabriseno

   on Jul 9, 2026

   Contributor

   More actions

   • We have the option to bypass the issue by reprojecting the SPICE file to a simple WCS with CDELT2=CDELT1 and PCi_j = Id before creating the JP2. This can be a quick fix for SPICE, but it is not very satisfactory on the long term

   This is what 2D Helioviewer expects in general, unless we add support on the server side. We have similar issues with 2D Solar Orbiter and GOES-19 because the source jp2s have a CROTA value which is also not handled in 2D mode because kakadu (program for tiling jpeg2000 files) doesn't support extracting out rotated tiles from the jp2, and we haven't yet invested the time to work around this.

   • The reprojection could be done server-side, but this adds some complexity and server load.

   If the source data needs to be reprojected at all, it must be either pre-computed in the jp2 provided to Helioviewer or computed server side to support screenshots and videos since those are generated server-side.

   • The web app could switch to the 3D mode when the metadata include CDELT1 ≠ CDELT2 or PCi_j ≠ Id

   Not ideal in the long term, but it could be done.

   • The 2D mode could be deprecated (maybe there are still issues with the 3D mode that prevent this?)

   Ultimately we do want the 3D view to be the main view, but it's not feature complete yet. It still needs support for events, screenshots, movies, and the VSO cutout.
7. ebuchlin commented on Sep 4, 2026

   

   ebuchlin

   on Sep 4, 2026

   More actions

   Thanks, I understand. If this is solved by #513 , can this one be closed?