raster calculator division. erroneous range of values?

1815
5
05-21-2019 12:32 AM
EricBriones
New Contributor

Raster division error?

Need help with raster calculator division

Raster A (CaptureA.PNG) and B (CaptureB.PNG) are both 32 bit. floating point

I wanted to divide A by B but the range of the values (CaptureAB.PNG) is erroneous (is it?)

Am i missing something here?

Thanks for you feedback.

0 Kudos
5 Replies
DanPatterson_Retired
MVP Emeritus
np.finfo(np.float32).min
-3.4028235e+38‍‍

numerator = np.arange(6000, 6300, 100, dtype=np.float32)

denominator = np.array([1e-12, 1e-11, 1e-10])

numerator/denominator
array([6.0e+15, 6.1e+14, 6.2e+13])

Not out of range.  It probably means that you have some values in the denominator that are extremely small.

To test, you need to report your minimum value for the denominator

EricBriones
New Contributor

Dan Patterson

Thanks for the feedback.

I attached the symbology of the rasters.

Jenks. 9 classes. Indeed, the denominator (KernelID_B) has extremely small values.

The quotient (KernelID_A/B) only has 3 classes.

Thanks a lot.

0 Kudos
DanPatterson_Retired
MVP Emeritus

Eric

Right-click on each raster, select Properties... Source, Raster Information, Band Metadata, then Statistics....

What is the min and max?

If you want to use symbology, you probably need to look at the data histograms to see where each dataset has values.

The range is possible given the ranges of values between the two

EricBriones
New Contributor

Hi Dan

Thank you very much. Really appreciate the help

KernelID A

min 0

max 6298.732......

KernelID B

min 0

max 0.579932.......

KernelID A/B

min 0

max 299913969664

I'm more interested on dividing Kernel A by Kernel B. then symbolize the the "Quotient"

Am i doing it wrong

Attched are the histograms

0 Kudos
DanPatterson_Retired
MVP Emeritus

Eric...

Huge number of zeros in both cases.  I hope 0 isn't nodata, if so, these should be dealt with  since they are throwing off the histograms, and if they are really nodata values, then that might explain why you haven't had any divide by 0 errors (?).

But in any event, you still have ranges in the histogram that can give rise to the values you found, so I don't things are mathematically going wrong. 

Perhaps they aren't what you expect, which mean you should address the questions surrounding the co-location that gives rise to the various values. 

Maybe it is time for a map or two and as for the ratio of the variables giving only 3 classes.... change it to something other than jenks so you get a real idea of what the values are in a different classification scheme.