Correlation between two (or more) layers?

[EstherLevy]

Hi,

I'm trying to quantify the relationship between layers, to see how well one feature (e.g. rainfall) correlates to another (e.g. land use). Is there a way to do this in arcgis?

Thanks
E

[KateParks]

Hi Esther,
I've used the geographically weighted regression tool to do this. There's a good tutorial on this in the resources centre (search for GWR or OLS) that explains the assumptions / limitations of using correlation on spatial data (issues of spatial autocorrelation etc) and shows you how to use the tools.  As far as I understand, both your datasets have to be shapefiles - it doesn't work on raster data.  I used zonal tools to aggregate my raster data to zones.
Hope this helps,
Kate.

[EstherLevy]

Hi Kate,

Thanks that does help, I think I'm sorted now 🙂

Esther

[ChrisBall]

Hi Esther,
I've used the geographically weighted regression tool to do this. There's a good tutorial on this in the resources centre (search for GWR or OLS) that explains the assumptions / limitations of using correlation on spatial data (issues of spatial autocorrelation etc) and shows you how to use the tools.  As far as I understand, both your datasets have to be shapefiles - it doesn't work on raster data.  I used zonal tools to aggregate my raster data to zones.
Hope this helps,
Kate.

Is there a method for doing this with nominal data, for example "habitat Type", where the habitat may be "Oak Woodland", "agriculture", etc.?  I know there are other methods for determining a relationship, but I'm losing where/how you do it in ArcGIS with data layers.

I thought I might be able to intersect point observations with a habitat layer, then use an attribute from the intersected habitat with the autocorrelation tool to determine if its clusterd or not, if it is, then there is a relationship between the variables. But the Auto-correlation tool also want's a numeric field.

If I classify each habitat type with a number value, would that yield real results?  Since in this case 2 is not 1 + 1, I think it may produce baloney. 

I have been getting summaries on habitat type and taking the top ranked classes, but I feel thier should be something else that is more appropriate (other than chi squared).

thanks

[ChrisBall]

Ok, I found what I was doing wrong.  I tried to edit/delete response to this thread but apparently not allowed to.

thanks

[lusekelokasunga]

Hi

Can you explain to me how you managed to do your correlation analysis-cause i also have quite a similar problem now with nominal data in ArcGIS. I am tryna correlate vegetation type with soil type. they are in shapefile format both are polygons.please help

[MatthewKing]

I have a "similar" question that I need to answer and I don't know what tool to use.  Specifically I want to see the relationship between soil type and vegetation type.  It's not a 1:1 relationship so I'm looking for some output that will basically say; Soil "A" contains XX% of veg type "B", XX% of veg type "C"....

Any thoughts on how to do this?

Thanks ahead for your answers.

[lusekelokasunga]

hi mat,

im sorry for writing on this old thread but i have exactly the same problem: i want to see the relationship between soil type and vegetation type and to quantify the significance of the correlation. with probably the same output as you stated above, of soil "A" contains XX% of veg type "B", etc. Please help me on this one.  

[ErinHuang]

I have the same question but GWR doesn't seem to work. It requires multiple fields in one layer. But I have two point layers, no field value required, just want to see if these two layers are spatially correlated. Is there a way in ArcGIS doing it? Thanks!

[JohnSulik1]

I have a "similar" question that I need to answer and I don't know what tool to use.  Specifically I want to see the relationship between soil type and vegetation type.  It's not a 1:1 relationship so I'm looking for some output that will basically say; Soil "A" contains XX% of veg type "B", XX% of veg type "C"....

Any thoughts on how to do this?

Thanks ahead for your answers.

It sounds like you want a contingency table (AKA crosstabulation). How you get there depends on whether your data is raster or vector.