topic Re: Correlation between Rasters? Raster Validation? in ArcGIS GeoStatistical Analyst Questions

Correlation between Rasters? Raster Validation?

KathrinSchulte-Braucks — Thu, 17 Feb 2011 14:33:05 GMT

Hey,

I have several raster files which I'd like to analyze concerning their level of similarity / correlation. What's the best way doing this? Apparently, Geostatistical Analyst only works with Vector Data. So do I have to apply the formula manualy with Map Algebra or is there a predefined tool e.g. for Pearson correlation or cross validation of raster files?

Thanks,

Kathrin

Re: Correlation between Rasters? Raster Validation?

EricKrause — Thu, 17 Feb 2011 15:30:33 GMT

I think you want the Band Collection Statistics tool in Spatial Analyst. It takes rasters as input and can output the correlation matrix (among other statistics). Just make sure to check the "Compute covariance and correlation matrices" box.

http://help.arcgis.com/en/arcgisdesktop/10.0/help/index.html#//009z000000p5000000.htm

Re: Correlation between Rasters? Raster Validation?

KathrinSchulte-Braucks — Fri, 18 Feb 2011 10:00:10 GMT

Hello Eric,

thank you very much for your answer, it really helped me a lot.

This is exactly what I was looking for!

Kathrin

Re: Correlation between Rasters? Raster Validation?

KathrinSchulte-Braucks — Wed, 02 Mar 2011 10:48:50 GMT

Hello,

is it somehow possible to calculate the statistical significance of the correlation values contained in the correlation matrix within ArcGIS?

Thank you very much,

Kathrin

Re: Correlation between Rasters? Raster Validation?

KathrinSchulte-Braucks — Sun, 06 Mar 2011 17:35:23 GMT

Any suggestions? Anyone? 😉

I'm comparing different raster files, some of them with different resolution. I'd like to find out how they correlate and how significant this correlation is.

Raster Band Statistics works so far, but I don't know how to compute the significance of the results with ArcGIS. Is it possible to calculate correlation and significance maybe with SPSS for grids?

Another problem occured: When comparing rasters with different cell size, some rasters get wrong MIN, MAX and STD values. So I first resize all grids to the cell size of the most detailed grid, and then compare, then the MIN, MAX and STD values are okay.

But since I have different resolutions, the amount of cells differs, so I can't just compare the VATs. I have somehow to account for the area. Would it helpful to resize the coarse grids to the same cell size as the grids with the best resolution and then compare the VATs with Excel? But how do I compute signifance and correlation in Excel or SPSS for a VAT which does not list each single value but only the count of each value?

Re: Correlation between Rasters? Raster Validation?

KathrinSchulte-Braucks — Mon, 07 Mar 2011 21:15:11 GMT

Okay, found my own solution:

Convert Rasters to Points, copy all Gridcode Columns into one table, open this table with R and calculate correlation coeff. and significance / regression in R. This works.

Re: Correlation between Rasters? Raster Validation?

EricKrause — Mon, 07 Mar 2011 21:18:36 GMT

You could also convert the rasters to points, join the points by location, and run the Ordinary Least Squares tool in Spatial statistics. Testing the significance of the coefficient in the linear model is the same as testing the correlation.

Re: Correlation between Rasters? Raster Validation?

KathrinSchulte-Braucks — Tue, 08 Mar 2011 11:55:55 GMT

Thank you again, Eric!

I'll try that out. I came across OLS before, but since I only have ArcView, I did not find the OLS Tool. Since I do have the Spatial Analyst Extension I do have access to OLS, when I uncheck "Hide Locked Tools". I somehow overlooked this in the Tool Help.

Re: Correlation between Rasters? Raster Validation?

CarlosSarmiento1 — Sun, 03 Apr 2011 17:56:02 GMT

Your conversation has helped me a lot. I have exactly the same problem as Katherine had. Thanks you both.

Re: Correlation between Rasters? Raster Validation?

FrankHanssen — Thu, 14 Apr 2011 07:00:41 GMT

You could also convert the rasters to points, join the points by location, and run the Ordinary Least Squares tool in Spatial statistics. Testing the significance of the coefficient in the linear model is the same as testing the correlation.

This seems to be the solution I need! I have modelled land surface clutter (as GRID) and I want to compare the modelled clutter GRID with the radar detected land cluttermap (GRID) from the same area. As I am a newbee to regression analysis I therefore need to clairfy some details.

I have converted both GRID`s to points and joined them using Spatial Join in my model like you described. I need to test the relationships in two ways:

* First: I want to compare the relationship between modeled values and radar detected values in the two datasets. When using OLS is it correct to use the modelled values as "Dependent variables" and the radar detected values as "Explanatory variables"?

* Seccond: I also want to perform a binary test to find the spatial relationship between modelled clutter (1= Present, 0= Non- present) and the radar detected values (1= Present, 0= Non- present). Can I use OLS also for this purpose or should I use a different tool?

Best regards

Frank Hanssen
GIS- analyst
NINA

Re: Correlation between Rasters? Raster Validation?

EricKrause — Thu, 14 Apr 2011 16:53:21 GMT

* First: I want to compare the relationship between modeled values and radar detected values in the two datasets. When using OLS is it correct to use the modelled values as "Dependent variables" and the radar detected values as "Explanatory variables"?

It doesn't matter which you call the dependent and explanatory variables. The p-value of the coefficient should be the same either way.

* Seccond: I also want to perform a binary test to find the spatial relationship between modelled clutter (1= Present, 0= Non- present) and the radar detected values (1= Present, 0= Non- present). Can I use OLS also for this purpose or should I use a different tool?

You can use the same tool to get the correlation, but you won't be able to trust the p-value.

We don't have a specific tool for it, but if you need to test the correlation of two binary variables, use the Phi Coefficient of Correlation. You can probably write a Python script to calculate and test it for you, but it won't be simple.

Re: Correlation between Rasters? Raster Validation?

SandraMelzner — Sat, 30 Apr 2011 12:09:14 GMT

I would like to compare as well two rasters. I tried to use Band collection statistics, but I can't open the resulting .asc in Arc Map!

I am quiet "virgin" in the statistical analysis of rasters! It could be that I need another function...??

My problem:
I have two raster with rockfall trajectories of different extent and different energy values. If you overlay these two rasters there are pixels,where eather one of the rasters has values OR there are pixels where both rasters have values, which are differnt from each other (these trajectories look a little bit like flow accumulation outputs)! So far I analized the data in substracting one from the other- differences are shown in different colours!
What kind of analysis could I use instead of this simple aproach?
Thanxxx 🙂

Re: Correlation between Rasters? Raster Validation?

KathrinSchulte-Braucks — Wed, 13 Jul 2011 21:11:45 GMT

@ ING2011: Open the resulting file with a text editor.

Re: Correlation between Rasters? Raster Validation?

PaulLeonard — Sat, 29 Dec 2012 22:04:17 GMT

Convert Rasters to Points,

Unfortunately this is not a viable option for large grids.

Re: Correlation between Rasters? Raster Validation?

Devon_Reid — Fri, 18 Mar 2016 11:40:43 GMT

Hallo Eric, I see this is from a looong time ago, however it is useful in what I'm doing now (unless there are better ways to test for collinearity now). I am preparing raster surfaces for a maxent model, and have calculated the correlation matrix. One question - is this Person's? It looks exactly like it for a group sample.
thanks,

Devon

Re: Correlation between Rasters? Raster Validation?

champikakariyawasam — Tue, 18 Sep 2018 04:30:52 GMT

I use high resolution 30s global asc raster layers for maxent modeling. I need to remove correlated variables, but none of the tools worked due to memory problems (exceed python limits). can somebody advise me pl? when i used 10m as files it worked well. Can i check the correlated variables using 10m data and select variables and go for my 30s data for modeling? can somebody instruct me pl?