Slice's EQUAL_AREA method

[StacieWolny]

Hi all -

I'm trying to use the Slice tool to group a script's result raster by a user-defined X%, which seems straightforward enough.  The resulting value distribution may not be normally distributed, which, the Slice doc says, is not optimal for use of the tool.  Now, I understand that using the EQUAL_INTERVAL method may produce empty classes, or classes with very few cells.  That makes sense because there just may not be cells with values that fall within the interval.  However, often when I use EQUAL_AREA I also get empty classes, which doesn't make sense to me, and the doc doesn't explain why this might happen.  It seems that if it's basing the grouping on area (which is always there) and not just value, there's no chance of missing values, unlike EQUAL_INTERVAL.  Can anyone shed light on why this would happen?  As it is, I might have to write my own Slice command to get better results.  Thanks!

~ Stacie

[curtvprice]

However, often when I use EQUAL_AREA I also get empty classes.  Can anyone shed light on why this would happen? 

Stacie - the problem you are running into here is probably having to do with the distribution of your data. Your data probably has a lot of tied values.

For example, consider a small 1 x 10 raster with just these cell values:

10 10 10 10 10 20 20 20 20 20

An equal area slice into two levels would look like this:

1  1  1  1  1  2  2  2  2  2

But what if we asked for three? We can't "split" these groups. So you will probably still get:

1  1  1  1  1 2  2  2  2  2

because of the ties.

[EricRice]

I pulled this directly from the Slice tool documentation under Usage.

"When using the EQUAL_AREA method, sometimes not all of the output zones  (classes) will have an equal, or even similar, number of cells (i.e.,  area). This may be an inherent result based on the nature of the input  values and the specified number of zones. If the results are deemed  undesirable, you can try using a fewer number of zones or applying a  statistics transformation (e.g., logarithm, square root, and so on) to  the input dataset."

Best,
Eric