Bug report #13995

singleband pseudocolor discrete rendered wrong

Added by William Kyngesburye over 4 years ago. Updated over 1 year ago.

Affected QGIS version:2.8.3 Regression?:No
Operating System:OS X Easy fix?:No
Pull Request or Patch supplied:No Resolution:end of life
Crashes QGIS or corrupts data:No Copied to github as #:22009


I'm trying to display discrete colors for ranges of values, but it's using values below what's set for a range. The raster (attached) is continuous float values from 0 to 2, with a 2 decimal precision (.01). I want breaks at .01, .02, and .1. My color table is set as:

0 white 0 to .01 (essentially 0)
0.01 light grey 0.01 to .02 (essentially 0.01)
0.02 med grey 0.02 to 0.1 (or 0.02 to 0.09)
0.1 dark grey 0.1 to 2
2 dark grey

See attached images for the correct and broken renderings.

Checking the raster values at the dark grey edge, I find that the .1 range really starts at .03, the next value up from .02 in the 2 decimal precision. The .02 range is OK presumably because for the precision it's the only value in that range.

I can workaround it by adding a .09 value with the same color as the .02 value.

I've tried scaling the raster by 100 so all values are integers, but this doesn't help.

raster.tif (1.38 MB) William Kyngesburye, 2015-12-15 02:23 PM

color_correct.png (7.17 KB) William Kyngesburye, 2015-12-15 02:23 PM

color_broken.png (7.36 KB) William Kyngesburye, 2015-12-15 02:23 PM

color_broken.txt Magnifier - color map (269 Bytes) William Kyngesburye, 2015-12-15 02:23 PM

color_workaround.txt Magnifier - color map for workaround (313 Bytes) William Kyngesburye, 2015-12-15 02:23 PM

Related issues

Related to QGIS Application - Bug report #14689: Singleband raster style rendering incorrectly/unexpectedly Closed 2016-04-14


#1 Updated by Piers van der Torren about 4 years ago

The problem here is that the behaviour of QGIS is different from what you expect. The behaviour is consistent, but the edges are defined different, like this:

0     white       <= 0
0.01  light grey  0 < .. <= .01 (essentially 0.01)
0.02  med grey    0.01 < .. <= .02 (essentially 0.02)
0.1   dark grey   0.02 < .. <= 0.1 (or 0.03 to 0.1)
2     dark grey   0.1 < .. <= 2 (or 0.11 to 2)

I've improved the automatic labels to make the ranges more clear, see https://github.com/qgis/QGIS/pull/2979

However, I noticed that the behaviour is not consistent with the gradient editor, which adds an extra class at the end, like you did.

If the type of breaks that you expected is also more common in other software, or otherwise more logical I think it makes sense to discuss if the renderer should be changed.

In the source documentation of QgsColorRampShader::exactColor() it says "Assigns the color of the lower class for every pixel between two class breaks.". That suggests the behaviour William expects actually also is the intended behaviour and the implementation is indeed wrong. Still the question is whether to fix the implementation or keep it and make it more clear what happens.

#2 Updated by Giovanni Manghi about 3 years ago

  • Easy fix? set to No
  • Regression? set to No

#3 Updated by Giovanni Manghi over 1 year ago

  • Resolution set to end of life
  • Status changed from Open to Closed

Also available in: Atom PDF