[ncl-talk] Bug (?) in filled contours since version NCL 6.5.

[Rick Brownrigg]

Hi Michael,

Several of us have taken a look at this issue and are able to replicate the
results with your script (Thanks for the nice clean example!). Oddly, the
bug occurs in very subtly different way between my Linux machine and my Mac
(the shape of the improper fill is the same, the colors are different).

I am afraid that finding a proper fix for this will take considerable
effort, and with our current staffing levels will, we will not be able to
address this in the immediate future. However, I may have a work-around
that might be acceptable.  I noticed that the viewport settings are pushed
right up against the top edge of the plot (its rare to see NCL plots like
this).  On a hunch, I offset it slightly interior to the plot, so:

   res at vpYF = 0.99999    ;; rather than 1.0

and the problem goes away.  One can not perceive the offset visually.-- it
looks flush with the top edge.

You might give this a shot to see if the incidence of intermittently
incorrect plots goes down.  I would be curious to hear what you find out.

I hope that helps...
Rick



On Mon, May 13, 2019 at 1:38 AM Michael Langguth <mlangguth at uni-bonn.de>
wrote:

> Dear NCL-community,
>
> we already posted this issue on the NCL GitHub repo (as
> #95 on 11th April), but would like to bring it to a
> probably broader attention via the ncl-talk. We hope not
> to annoy some of you with a double posting, but we still
> could not solve the problem and would like to update to
> the newest NCL version in the near future.
>
> Starting with NCL 6.5.0, a random-like bug in the filled
> contour plots being produced routinely at our institute
> (Institute of Geoscience - Department Meteorology of the
> University of Bonn) arose.
>
> We plot GFS data obtained from the NCEP data server over
> a domain spanning from the West Atlantic to Eastern
> Europe, i.e. our plot domain crosses the prime meridian.
>   While we had never problems using NCL 6.3.0, the
> contours sometimes aren't filled correctly neither with
> NCL 6.5.0 nor with NCL 6.6.2.
>   More specifically, uniformly colored segments over
> several levels (which should have different fill colors)
> are drawn. These segments are likely to be confined by
> the prime meridian.
>   The problem seems to occur randomly since a small change
> in the plotting domain may yield a correct visualization
> and since the plotting process may succeed using slightly
> different data sets (i.e. another GFS data file).
>
> For a visualisation of this issue we would like to refer
> to our GitHub post: https://github.com/NCAR/ncl/issues/95
>
>   The exemplary script and the data set (cropped data of
> GFS run 2019/01/23 00 UTC +087 forecast) can be accessed
> from:
>
>   link to script:
> https://uni-bonn.sciebo.de/s/I9dXjEAz8twDAtX
>   link to data :
> https://uni-bonn.sciebo.de/s/5lctA71qie3TSEp
>
>   Our system runs on Linux 4.12.14 (OpenSuse 15.0) and NCL
> was installed by using the binary from NCAR.
>
>   Do you have any ideas how to fix this problem or is it
> related to a deep bug in the NCL plotting routine?
>
>   Many thanks in advance and kind regards!
>   M. Langguth and A.Demers
>
>
> -------------------------------
> ------------------------------------
> M. Sc. Michael Langguth
> Institute for Geoscience - Department Meteorology
> University of Bonn
> Auf dem Hügel 20, Bonn
> phone: +49 228 73 5186
> email: mlangguth at uni-bonn.de
> _______________________________________________
> ncl-talk mailing list
> ncl-talk at ucar.edu
> List instructions, subscriber options, unsubscribe:
> http://mailman.ucar.edu/mailman/listinfo/ncl-talk
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.ucar.edu/pipermail/ncl-talk/attachments/20190514/40363a7b/attachment.html>