topic OLS or GWR regression on rasters in Spatial Statistics Questions

OLS or GWR regression on rasters

JeremyProville — Fri, 28 Oct 2011 15:19:50 GMT

Hi,

I've got two raster layers to do with a specific crop:
Yields
Suitability Index for growing

What I'd like to do is use the suitability layer, which covers the entire globe, to impute the potential (expected) yield in places where the crop is not grown. I would do this using the actual yields layer.
I can't use rasters as inputs for the OLS or GWR tool. I tried converting them to vector format, doing a spatial join, and then running a regression but it won't work as the datasets are simply too big when I convert them to polygons.

How would I run a regression on rasters, to impute the expected values of one based on another?

Thanks!

Re: OLS or GWR regression on rasters

ChrisBater — Fri, 28 Oct 2011 17:14:20 GMT

Can you just run your regression on a subset of your data? Take a systematic sample of your rasters and run your anayses on those.

Re: OLS or GWR regression on rasters

JeremyProville — Tue, 01 Nov 2011 20:21:00 GMT

Can you just run your regression on a subset of your data? Take a systematic sample of your rasters and run your anayses on those.

Hi passiveoptical,

Thanks, your suggestion is a good idea and I'm sure would have been a good solution in other circumstances. I've been running into problems when trying that approach: since the dataset is global in scale and only one of many regressions I'll have to do, the sampling would be very time intensive. I suppose I could script it but there's another issue in that after getting the expected values, I have to reconvert to raster. The regression outputs need to be combined with other datasets subsequently (using the raster calculator).

I might just have to try the batch scripting method if there's no other ideas...I do find it odd that the regression tools don't accept rasters of identical dimensions though, it seems like a pretty straightforward and necessary operation.

Thanks for the advice!