contourf and pcolormesh can look bad in pdf viewers, with lots
of extra "gridlines" or lines around the boundaries of the contours.
This is an example contourf PDF converted to png at 100 dpi to mimic what a screen would show.
Note that this is just an issue with the PDF viewers. Adobe Acrobat has an option to turn off "Smooth line art", and doing so removes the anti-aliasing and hence the white lines at the contour boundaries.
See this comment_.
.._comment: matplotlib/matplotlib#9574 (comment)
The same plot rendered by Matplotlib into a 100-dpi png does not have the extra whitish dashed lines around each contour
This is an example contourf directly printed to a 100-dpi png.
The reason the pdf file renders like this is because the pdf rasterizer will often employ anti-aliasing be default. Software when displaying the PDF will rasterize the polygons that make up the contours and apply anti-aliasing to their edges. However, this is usually done per-polygon, so the first polygon with anti-aliasing doesn't know that it is going to be beside another polygon, so it anti-aliases with the white background of the figure. The adjacent polygon does the same, and hence the pixels at the boudnary between the two polygons are lightened. The same figure rendered with a black background yields dark lines between the contours.
This is an example contourf (converted to png at 100 dpi to mimic what a screen would show) but with the background black. Note that contour interfaces are now dark.
Note that at a fundamental level, this is not Matplotlib's fault. If you rasterize the PDF produced by matplotlib at 300 dpi (i.e. using Imagemagick or another rasterizing tool), the anti-aliasing is not as noticeable. If you could turn anti-aliasing off for your PDF viewer, this problem would not occur.
This is an example contourf PDF converted to png at 300 dpi.
What should be done depends on your application.
If you are producing PDF
files, setting the edgecolor='face' option in contourf and setting
the linewidths=0.2 yields a good result on a 200-dpi monitor. Of course
this means that the underlying polygons are now overlapping by 0.07 mm.
That is not too likely a problem for normal viewing, but if contours are
being zoomed in on to a very high level that will cause some distortion of the
exact boundaries.
If you are using alpha values in your contours, then there is a problem with the overlapping, because now your alpha value is doubled, likely over 1.0.
Countourf at different values of edgecolors and linewidths in a PDF
converted to 100 dpi png for display here. Original pdf.
Here, the right answer is to almost always just use edgecolors='none'.
Note, however, the poor performance when alpha is less than 1.
Countourf at different values of edgecolors and linewidths in a PNG
rendered at 100 dpi by Matplotlib Original png.