Geometrical interval classification

[StefaniaStevenazzi]

I've tried to reclassify a raster map (floating) using geometrical interval. I've tried two ways: the first one using the Reclassify tool and the second one using the classification window in Properties->Symbology. The results are sometimes fairly different.

Does someone know the reason?

Thank you.

[DanPatterson_Retired]

Can you provide an example comparison so that people don't have to try it themselves?

[StefaniaStevenazzi]

Here are two example. The two classifications have been done on the same raster, but the break values (and the statistics) are different. The raster has been obtained applying a math equation (Single Output Map Algebra) using different rasters with the same extension. And a mask is imposed in the Environment settings.

Could these options produce the mistake?

1) Classification from Reclassify Tool -> Classify

Geometrical interval, 5 classes

Class 1: 0.000002443 ≤ values ≤ 0.000003101

Class 2: 0.000003101 < values ≤ 0.000007076

Class 3: 0.000007076 < values ≤ 0.000031095

Class 4: 0.000031095 < values ≤ 0.00017624

Class 5: 0.00017624 < values ≤ 0.001053348

Classification Statistics

Count: 256042

Min: 0.000002443

Max: 0.001053348

Sum: 9.910332135

Mean: 0.000038076

STD: 0.000070099

2) Classification from Layer properties -> Simbology -> Classified -> Classify

Geometrical interval, 5 classes

Class 1: 0.000002443 ≤ values ≤ 0.000015756

Class 2: 0.000015756 < values ≤ 0.000017364

Class 3: 0.000017364 < values ≤ 0.000030677

Class 4: 0.000030677 < values ≤ 0.000140893

Class 5: 0.000140893 < values ≤ 0.001053348

Classification Statistics

Count: 256042

Min: 0.000002443

Max: 0.001053348

Sum: 10.44167238

Mean: 0.000040781

STD: 0.000070162

[DanPatterson_Retired]

You indicate a mask was used...perhaps one of the options doesn't support a mask.  You could try to rerun the analysis without the mask being specified in order to rule out this possibility

UPDATE
I just found a reference in Eric Rice's post in this link on a slightly unrelated issue suggesting that not all values are used in calculating statistics via symbology, unfortunately the link to the article doesn't work.

[StefaniaStevenazzi]

I've tried as you suggest. Indeed the two classifications are the same. Thanks for the advice.

I've also read the post that you suggest.

[DanPatterson_Retired]

Not sure if that means that the mask is the cause of the problem or not or you get the same results with or without the mask?

[StefaniaStevenazzi]

With the mask I've obtained the result in the examples above (with two different classifications using the two methods), while whithout the mask I've obtained the same classification (that is the same break values) using the two methods. Without the mask the break values are the same obtained with Reclass tool in the case with the mask (example n. 1).

This means that the mask has some influence on the classification.