topic Re: Correlation between DEM, slope, and aspect in ArcGIS Spatial Analyst Questions

Correlation between DEM, slope, and aspect

EvanEllicott — Wed, 14 Mar 2018 17:07:55 GMT

I am building a suitability regression for a tree plantation. Like many things related to flora, elevation, slope, and aspect are important variables for growth and thus site selection. However, I am trying to wrap my head around how each of these can be used if 2 (slope and aspect) are derived from the 3rd (DEM). Is this analytically acceptable? Is there a way to account for the correlation and the error which must propagate?

Any advice, opinion and insight is welcome.

Re: Correlation between DEM, slope, and aspect

MervynLotter — Wed, 14 Mar 2018 19:43:14 GMT

To assess the correlation between these 3 rasters, you can use the Band Collection Statistics tool. See Band Collection Statistics—Help | ArcGIS Desktop and How Band Collection Statistics works—Help | ArcGIS Desktop

Re: Correlation between DEM, slope, and aspect

Anonymous User — Mon, 20 Aug 2018 16:57:03 GMT

Hi Evan,

Slope and Aspect are acceptable products of a DEM to use in analysis, as least in my experience. If you'd like to read a bit more about how they're calculated, there's a great explanation for each:

How Slope works—Help | ArcGIS Desktop

How Aspect works—Help | ArcGIS Desktop

What are your concerns about errors and distortion? It's true any inaccuracies in the original DEM will propagate through into the Slope and Aspect rasters, but then I would recommend either trying to obtain more accurate DEM data or filling Sinks, etc. But maybe we can clarify your specific concerns?

Re: Correlation between DEM, slope, and aspect

DanPatterson_Retired — Mon, 20 Aug 2018 22:34:47 GMT

Just drop the word "correlation"... slope angle does not depend on slope aspect. The other two are derived from the dem.

What you are looking for is an "association" between you variable(s) of interest and other variables that may influence it.

As soon as you say "correlation" you are implying that there is a numeric association between variables that is predictable. Many things in nature do not operate on that interval/ratio scale but may be associated in the sense that "if the slope faces south... life is good" not "if the slope is 10% you will get a yield of X" or other similar terms.

Overlay mapping and simple things like chi-square testing should be your first go to unless you have a compelling case to use parametric statistics