[Qgis-user] Raster Calculator bug?

Fri Jul 31 08:31:11 PDT 2015

Hi Stefan,

> On Jul 31, 2015, at 9:37 AM, Stefan Kiefer <st_kiefer at web.de> wrote:
> 
> Hi Nick,
> back to office I was eager to try by myself. Actually it seems that the result of multiple AND or multiple layers - I didn't check this by now - results in values slightly lower 1 (e.g. 0.9995 in my case). And therefore maybe rounded to "0". What I have done is the folowing:
> 

Thanks! I did try your suggestion and it provided me the results I was seeking of only showing the red routes:

(1/("map at 1" < 238 AND "map at 1" > 210)) + (1/("map at 2" < 123 AND "map at 2" > 94)) + (1/("map at 3" < 130 AND "map at 3" > 98))

Perhaps my math is fuzzy.  Why would the divisor resolve this situation whereas multiplier in following has problems?

((“m at 1" < 238 AND “m at 1" > 210) * 1) * ((“m at 2" < 123 AND “m at 2" > 94) * 1) * ((“m at 3" < 130 AND “m at 3" > 98) * 1)

The floating point values is a why I believe we have a rounding bug in Raster Calculator.  I suspect that during the calculation a number is converted to floating point, between 0 and 1 (eg 0.9995), and that causes the equation to result in all 0s.  Its unusual that the calculator is performing floating point math because these are all integer values being operated on by value comparison operators and other integer numbers without any division which would cause floating point conversion or multiplication by a floating point.  The numbers are all whole integers.  I was hoping the original author could chime in on the tool and perhaps this is a bug.

I’m on Mac OSX using 2.10.1 release and see same results on 2.8 release.

I’m suspecting this may be a bug related to:

Raster calculator always returns float32 tiffs
https://hub.qgis.org/issues/10965 <https://hub.qgis.org/issues/10965>
Raster calculator produces only 0 values with conditional expression
https://hub.qgis.org/issues/11682 <https://hub.qgis.org/issues/11682>

There is this suggestion on the mailing list of how to resolve, however am a bit confused on the methods.  The workaround may be to reproject the raster layer and I’m unsure why that’s required:
http://lists.osgeo.org/pipermail/qgis-user/2014-November/029873.html <http://lists.osgeo.org/pipermail/qgis-user/2014-November/029873.html>
http://lists.osgeo.org/pipermail/qgis-user/2014-November/029878.html <http://lists.osgeo.org/pipermail/qgis-user/2014-November/029878.html>

Thanks again,
Nick

> (1/("pm2-5-europe-2001-2010 at 1">240 AND "pm2-5-europe-2001-2010 at 1"<250))*(1/("pm2-5-europe-2001-2010 at 2">139 AND "pm2-5-europe-2001-2010 at 2"<145)) * (1/("pm2-5-europe-2001-2010 at 3">80 AND "pm2-5-europe-2001-2010 at 3"<85))
> 
> which results in a perfektly fitting mask of my pseudo demand.
> Mayby you could verfy this with your data
> 
> cheers
> 
> Stefan
> 
> Am 31.07.2015 um 08:48 schrieb Stefan Kiefer:
>> Hi Nick,
>> you are absolutely right. My thought was, that you get A layer with distinct values to identify the road. For a mask you are on the right way, and I either don't understand the behaviour except that you operate over three layers, which of course should work.
>> Have you tryed to generate a composit of the three layers and mask the single values resulting for road structures? (it's more or less what I expected from my first approach.)
>>  
>> cheers
>>  
>> Stefan
>> 
>> > Nick Papadonis <npapadonis at gmail.com> <mailto:npapadonis at gmail.com> hat am 31. Juli 2015 um 08:31 geschrieben:
>> > 
>> > 
>> > Hi Stefan,
>> > 
>> > It’s my understanding black has a value of 0 in the resulting layer.
>> > 
>> > I tried this and it results in similar image to step (a) and also includes other colors at lower intensities mixed in with the red. The red has the highest intensity in the greyscale. I’m looking to create a binary image with just the colors of red in the palette I choose and using this trace vectors over the paths.
>> > 
>> > Thanks,
>> > Nick
>> > 
>> > > On Jul 31, 2015, at 2:04 AM, Stefan Kiefer <st_kiefer at web.de> <mailto:st_kiefer at web.de> wrote:
>> > > 
>> > > Hi Nick,
>> > > I believe it is black bcause you always get a value of "1". Unfortunately I can not verify this, because I have no QGis by this moment. Most propably you wanted to calculate:
>> > > 
>> > > (“m at 1" < 238 AND “m at 1" > 213 AND “m at 2" < 123 AND “m at 2" > 98 AND “m at 3" < 125 AND “m at 3” > 99) * ((“m at 1" < 238 AND “m at 1" > 210) * "m at 1") + ((“m at 2" < 123 AND “m at 2" > 94) * “m at 2") + ((“m at 3" < 130 AND “m at 3" > 98) *“m at 3"))
>> > > 
>> > > cheers
>> > > 
>> > > Stefan
>> > > 
>> > > > Nick Papadonis <npapadonis at gmail.com> <mailto:npapadonis at gmail.com> hat am 31. Juli 2015 um 07:49 geschrieben:
>> > > > 
>> > > > 
>> > > > One more comment. The resulting layer histogram is showing the pixel range spread over frequency in floating point values. Is the raster calculator performing floating point math with potential rounding error?
>> > > > 
>> > > > I found it also interesting that the following expression resulted in a layer, which when inspected for band values, has integer values of 2 and 3. 3 being the value I want for the red route.
>> > > > 
>> > > > a) ((“m at 1" < 238 AND “m at 1" > 210) * 1) + ((“m at 2" < 123 AND “m at 2" > 94) * 1) + ((“m at 3" < 130 AND “m at 3" > 98) * 1) 
>> > > > 
>> > > > I then change the expression to only use values 2 and greater and this shows properly:
>> > > > b) ((“m at 1" < 238 AND “m at 1" > 210) * 1) + ((“m at 2" < 123 AND “m at 2" > 94) * 1) + ((“m at 3" < 130 AND “m at 3" > 98) * 1) > 2
>> > > > 
>> > > > I then changed the expression to ensure all three values are obtained and it results in a black image of 0’s. I was expecting only the red route to appear as it resulted in value of 3 in step (a). 
>> > > > 
>> > > > ((“m at 1" < 238 AND “m at 1" > 210) * 1) + ((“m at 2" < 123 AND “m at 2" > 94) * 1) + ((“m at 3" < 130 AND “m at 3" > 98) * 1) > 2.1
>> > > > ((“m at 1" < 238 AND “m at 1" > 210) * 1) + ((“m at 2" < 123 AND “m at 2" > 94) * 1) + ((“m at 3" < 130 AND “m at 3" > 98) * 1) >= 3
>> > > > 
>> > > > I’m wondering how much testing the Raster Calculator has gone through and if there is a possible bug here. Perhaps something to do with floating point?
>> > > > 
>> > > > Thanks again
>> > > > 
>> > > > > On Jul 31, 2015, at 12:39 AM, Nick Papadonis <npapadonis at gmail.com> <mailto:npapadonis at gmail.com> wrote:
>> > > > > 
>> > > > > Folks,
>> > > > > 
>> > > > > I’m using QGIS 10.1. The following expressions result in a black raster of 0’s, when I expected only red pixels to appears in the binary image indicating routes on a map:
>> > > > > 
>> > > > > a) (“m at 1" < 238 AND “m at 1" > 213 AND “m at 2" < 123 AND “m at 2" > 98 AND “m at 3" < 125 AND “m at 3” > 99) * 1
>> > > > > b) ((“m at 1" < 238 AND “m at 1" > 210) * 1) * ((“m at 2" < 123 AND “m at 2" > 94) * 1) * ((“m at 3" < 130 AND “m at 3" > 98) * 1)
>> > > > > 
>> > > > > I then tried the following individual expressions for each band as separate steps (sanity check) and they work to cover the pixels in range:
>> > > > > c) (“m at 1" < 238 AND “m at 1" > 213) * 1
>> > > > > d) (“m at 2" < 123 AND “m at 2" > 98) * 1
>> > > > > e) (“m at 3" < 125 AND “m at 3” > 99) * 1
>> > > > > 
>> > > > > I then tried the following expression which appears to create a proper greyscale image focusing on the red pixels. I replaced the multiplication with addition to see what was happening:
>> > > > > f) ((“m at 1" < 238 AND “m at 1" > 210) * 1) + ((“m at 2" < 123 AND “m at 2" > 94) * 1) + ((“m at 3" < 130 AND “m at 3" > 98) * 1)
>> > > > > 
>> > > > > The resulting raster has a Min = 0 and Max = 1.998. I was expecting it to be Min = 0 and Max = 3. The value of 3 would indicate all 3 bands were positive on color match. I then go to the layer properties and load calculate min/max again and it is Min = 0 and Max = 3. I tried to change the min/max settings on they layer and these settings will not stay set. The layer goes back to Max = 1.998. What’s even more odd is the max being a floating point number. I suspect that may be part of the issue. Anyone know why this is the case for integer band values? Has anyone successfully used the Raster Calculator to perform this sort of work before?
>> > > > > 
>> > > > > Thanks again,
>> > > > > Nick
>> > > > 
>> > > > _______________________________________________
>> > > > Qgis-user mailing list
>> > > > Qgis-user at lists.osgeo.org <mailto:Qgis-user at lists.osgeo.org>
>> > > > http://lists.osgeo.org/mailman/listinfo/qgis-user <http://lists.osgeo.org/mailman/listinfo/qgis-user>
>> >
>> 
>> 
>> _______________________________________________
>> Qgis-user mailing list
>> Qgis-user at lists.osgeo.org <mailto:Qgis-user at lists.osgeo.org>
>> http://lists.osgeo.org/mailman/listinfo/qgis-user <http://lists.osgeo.org/mailman/listinfo/qgis-user>
> _______________________________________________
> Qgis-user mailing list
> Qgis-user at lists.osgeo.org
> http://lists.osgeo.org/mailman/listinfo/qgis-user

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.osgeo.org/pipermail/qgis-user/attachments/20150731/760f0cf2/attachment.html>