You might categorize the continuos variable (density) with Reclasiffy command (natural jenks, geometric intervals, quantiles), and then isolate the interest class

You can also use the contour tool

Hey John,

as Miguel proposed, the contour tool seems to be the right tool for you. You'll find it in Toolbox->Spatial Analyst Tools->Surface-> Countor (create contourf for specified equal interval) or alernatively Contour list (where you can define the contour you need (e.g. if they should not have a specific equal interval)).

This presumes, of course, that you have converted the density map generated by kernel density tool (units per square area!) into percentage values you need, since you asking for 95% threshold. In particular that is quite easy to do using Map Algebra->Raster Calculator, if the 95% refers to the normalized density value.

Best regards,

Jewgenij

unfortunately reclassify + quantile returns break values based on the count of cells => any class will have the same number of cells (area based quantiles) ... instead of: any class will have the same number of e.g. population.

by the way - if you want to use the area-based approach within model builder you should take "slice" (instead of "reclass") - this one recalcs the break values every time you run the model.

for isopleth creation i've just found one tool --- geospatila modelling environment (GME)

and i am still looking for alternatives within arcgis

... @ jewgenij: how do you calc a normalized density raster by means of raster calculator which returns aggregated percentage of the VALUE

best robert

just have designed a model builder model which performs that task ...

@Robert: in general, nice job! The approach you propose is interesting finger exercise but I think (without that I have run such a model) a bit slow performing due to the introduction of the VATs. In terms of automation, the task can be solved much faster introducing a small python script for the cumulative raster sum instead of converting a float raster to integer,  and computing the cumulative sum within the VAT. In my experience, the attribute tables in ArcGIS are quite havy objects. As for me, meanwhile I try to avoid to calculate within attribute tables for raster related tasks. Further, if you ask me, I would completely work in python scripts without toutching Model Builder at all :-)

In terms of laziness, the value of the raster sum can also be easily derived just by reading the classification statistics of the raster (Layer properties-> Symbology-> Classified -> Classify button (here, in the upper right corner, the classification statistics is provided). I have no clue why this method was not introduced in the regular raster statistics (as did for MAX, MIN, etc.) because this is realy not difficult to obtain and obviously works on the fly as a part of the common ArcGIS application (this could then indeed be simply implemented in the Model builder). Having this value the entire task (except the kernel density estimation of course) can be easily performed in the raster calculator. 

It  depends on what you want to normalize: just  the value range (using Maximum raster value - in this case the value range is 0-1) or the raster dataset by the cumulative sum of the raster (in this case the raster sum is 1).

Some remarcs regarding your first message: Quantil classification fullfil the purpose for which it was introduced. Namely to describe the position within a distribution, as you correctly mentioned, based on the cell count not the value range.  Regarding this point I did not understand the disappointment you expressed. If you want to have classes with equal value range you can easily apply equal interval in the reclass tool. The slice tool that you propose to use, does not behave much different, if you specify the equal interval method. In the same time, the equal area method of the slice tool is exactly the classification using quantils, and slicing with natural jenks is nothing else than the reclass with natural jenks. For that matter, I see no much difference in application of the one or the other tool other than personal preference.

Best regards,

Jewgenij

Jewgenij Torizin

I'm running into the same problem as OP. Quite a few years have passed, but I hope you can help me out. You posted the text below and I would like to do just that, but I don't know how. Could you possibly explain how you would get your Kernel Density into percentage using Raster Calculator?

"This presumes, of course, that you have converted the density map generated by kernel density tool (units per square area!) into percentage values you need, since you asking for 95% threshold. In particular that is quite easy to do using Map Algebra->Raster Calculator, if the 95% refers to the normalized density value."

Many thanks!