Kernel Density Interpretation and Valid Statistical Comparison

[GarrettRSmith]

Hello All,

I have two kernel density rasters obtained from two different point layers, representing two distinct phenomenon. I am trying to see if there is statistically significant correlation between the location and intensity of these densities between the two point layers.

I have two questions.

First, how should a kernel density output be interpreted? I understand that the higher values represent more dense areas. However, how am I supposed to interpret the numerical values that are present in my raster dataset? In one example my values range from 0 to 10,244.214. There are only 75 points for this density analysis so I am not sure what the 10,244 refers to since there are no cells that would have that many values within it?

Second, does anyone know of a statistically valid method to compare the two density rasters to see if they are significantly different from one another? Should the data be standardized before running this analysis? I have found the Jaccardi index that have been used to compare rasters, but not sure if there is anything else out there that someone would recommend for kernel density comparisons?

Thanks for any suggestions or interpretation help.

[DanPatterson]

How Kernel Density works—ArcGIS Pro | Documentation

Understanding density analysis—ArcGIS Pro | Documentation

what are the two distinct phenomena?

is there an expectation that you can compare the distributions?

the appropriateness of any test will depend on the measurement scale and type of data you have

[GarrettRSmith]

The phenomena are recreation crowding and recreation conflict. You might expect that as crowding rises so does conflict.

Users were able to place points along recreation trails where they have experienced either crowding or have been involved in a conflict with another recreational user.

Running kernel density of each of the point layers returns different values, so that is why I was thinking that I should standardize the values in some way. In order to standardize I would really like to know what the output values mean. 

Maybe I should use point density analysis instead of kernel density since that comparison is more straight forward when using points of incidents?

[Anonymous User · Accepted solution]

For a statistical analysis of two rasters, it would depend on what you want. Raster Calculator can do pretty much any equation to one or more rasters and generate an output based on those values. The only prep I have found to be helpful in that regard is making sure that the rasters are snapped to each other, using the snap raster setting.

[GarrettRSmith]

Thanks for the suggestion. I used Raster Calculator to perform a Jaccard Similarity analysis between the two layers.