[ncl-talk] contour plot efficiency for unstructured data and some issues.

Yu, Hao-Cheng danishyo at gmail.com
Thu Feb 27 19:15:37 MST 2020


Hi everyone

I'm using latest precompile v6.6.2 ncl to plot with unstructured data.
sfXarray, sfYarray and sfElementNodes were used to do the contour.
Things are going well with FE mesh under 600~700k nodes (about 1.2 million
elements).
However, if mesh increase to certain amount, say 1.8 million nodes (5
million elements), I have to use multi-threads with rasterfill to get
reasonable efficiency, but the figure output is totally not right.
I have no chance to test areafill because after run few hours, no any
contour plot can be generated.

I also try with Matlab, and get the correct contour plot.
The attach figures show the difference.
Matlab took about 90s to produce the plot with 1 core.
ncl took 130s under 6 threads,  20 mins with 1 thread.

>From ncl website, I saw MPAS example with 2.6 million cell and get quite
reasonable efficiency with rasterfill.
I notice sfElementNodes is not used in script.
When I comment out this resource, ncl can produce the plot about 60s with 6
threads, but without element table, the plot is still wrong.

Just curious, will it be any chance to get better efficiency and produce
correct figure on PyNGL?

Dan
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.ucar.edu/pipermail/ncl-talk/attachments/20200227/0c0217e7/attachment-0001.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: matlab.PNG
Type: image/png
Size: 36327 bytes
Desc: not available
URL: <http://mailman.ucar.edu/pipermail/ncl-talk/attachments/20200227/0c0217e7/attachment-0002.png>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: ncl.PNG
Type: image/png
Size: 328486 bytes
Desc: not available
URL: <http://mailman.ucar.edu/pipermail/ncl-talk/attachments/20200227/0c0217e7/attachment-0003.png>


More information about the ncl-talk mailing list