Linear Regression Raster data

[MadhusankaJayawardhana]

I have a five raster layers with same cell size and same spatial reference. I want to perform regression analyze to find out the relationship between these layers.  How do I perform a regression analyze in ArcGIS? It hard to export as ASCII file because these raster images have 5000x5000 grids. 

[DanPatterson_Retired]

Geographically weighted regression... from the help files

[NeilAyres]

If you wanted to do simple regression, you could create a subset using :

Create Random Points—Help | ArcGIS Desktop 

and

Sample—Help | ArcGIS Desktop 

Then just process the output in Excel or something.

[DanPatterson_Retired]

But in any event... heed the warning in the topic link I posted.  And of course you will need data measured on an interval/ratio scale since nominal or ordinal data are not appropriate for any sort of regression analysis and you would be better performing tests of association (perhaps Chi square)

• You should always begin regression analysis with Ordinary Least Squares (OLS) regression. First find a properly specified OLS model, then use the same explanatory variables to run GWR (excluding any "dummy" explanatory variables representing different spatial regimes).

[MadhusankaJayawardhana]

GWR regression works only for the feature classes. But I have a raster datasets. How do I perform GWR analyze using rasters.

[DanPatterson_Retired]

Short of converting back to vector... what are your variables? are they floating point rasters of data on an interval or ratio scale? or are they nominal/categorical data?

[MadhusankaJayawardhana]

There are four floating point rasters and one categorical raster. The attributes are rooftop surface temperature, building height, rooftop area, rooftop slope,  and albedo. Rooftop aspect layer has a categorical data. These all rasters have the same spatial resolution of 0.25m. 

[DanPatterson_Retired]

why not classify the rasters into categories and use Combine to produce a map showing the various combinations.  These data seem more appropriate to a test of association/difference than a correlation/regression.  (ie south-facing, dark surfaces will have a higher temperature, than north-facing light surfaces etc etc)  I don't think you are trying to predict anything or describe the regime by an equation are you?

[MadhusankaJayawardhana]

I performed raster based multiple regression with Clark Labs Terrsat (Idrisi) software to find the correlation between above-mentioned layers. Thank you for all

[MadhusankaJayawardhana]

I performed raster based multiple regression with Clark Labs Terrsat (Idrisi) software to find the correlation between above-mentioned layers. Thank you for all