topic Re: Combine on two rasters with different cell size in ArcGIS Spatial Analyst Questions

Combine on two rasters with different cell size

DitlevReventlow — Wed, 15 Jul 2015 19:54:49 GMT

Hey everyone.

I have used the tool combine on two datasets that did not have the same spatial resolution. I can see that "Combine" has chosen the bigger cell size as the output raster spatial resolution.

How has combine calculated the output raster value for the raster that had the small cell size? I cant see anything about it in the combine tool help.

Thank you very much.

Best regards

Ditlev Reventlow

Re: Combine on two rasters with different cell size

IanMurray — Wed, 15 Jul 2015 20:05:01 GMT

Hi Ditlev,

I'm not sure which resampling method it would use(I would assume Nearest Neighbor, but ArcGIS has several other resampling options, including Majority, Bilinear and Cubic, and as you said there is no tool documentation for which is being used). If you want to guarantee which method is being used, I would recommend Resampling the smaller resolution raster first to the size of the larger resolution with the resampling method of your choice, then combining your rasters.

ArcGIS Help 10.1

Re: Combine on two rasters with different cell size

DitlevReventlow — Wed, 15 Jul 2015 20:31:45 GMT

Thank you for this Ian, The problem is that I have already done the analysis, so I have to write the consequences and implications of this. I guess though that a solution could be to try the different resampling methods on the finer raster and see if one of them gives the same result as combine.

Re: Combine on two rasters with different cell size

DitlevReventlow — Wed, 15 Jul 2015 20:43:24 GMT

Nope, this doesn't work, because although the cell size is the same, the position of the cell in the "new resampled file" and the combined file is different. Therefore, it will not be the same smaller cells that the two coarser raster datasets are derived from. Hope there is another solution to solving this problem.

Re: Combine on two rasters with different cell size

SteveLynch — Wed, 15 Jul 2015 20:50:07 GMT

read up - processing extent and snap raster

Re: Combine on two rasters with different cell size

DanPatterson_Retired — Wed, 15 Jul 2015 22:19:49 GMT

Follow Steve Lynch's advise and when you are using tools in ArcToolbox, there is the oft-ignored Environments button in the lower right of the dialog. That is where you should set all your raster parameters prior to running any tool. Do not assume that once you have set something ... OR ... you have set some of them somewhere else, that they will be retained and/or used. The rule is...if it isn't set, then a default will be used and a default is only as good as your data are perfect. You need to specify the extent, cell size and snap raster to align the cell edges...explicitly.

Re: Combine on two rasters with different cell size

DitlevReventlow — Thu, 16 Jul 2015 07:51:19 GMT

Great, thank you Dan and Steve. I tried all the different resampling method, except form majority, and none of them created the same number.

The reason why I havent tested the majority method is because when I do the resampling with majority there is no output raster afterwards. I have tried it several times and each time the same. How can this be that majority doesn't work like the other resampling methods?

Also Is it possible that the "combine" tool uses another resampling method than the 4 options in the "resample" tool?

Re: Combine on two rasters with different cell size

DanPatterson_Retired — Thu, 16 Jul 2015 08:31:26 GMT

Why not just go with the cell size of the finer grid by setting the cell size, Cell Size (Environment setting)—Help | ArcGIS for Desktop , the extent and snap raster as suggested by Steve. Then do the combine. There is no interpolation, no aggregation ... just the bigger cell sized raster gets chopped into smaller bits to match the smaller cell sized raster... Of course this recommendation is contingent upon knowing what you are working with and what you intend to do it since ANY kind of combine operation interpretation is contingent upon a full understanding of the inputs and the processes that they were subjected to... And screen grabs of what you are combining would help since the spatial arrangement of the 'things' may prompt further recommendations.

Re: Combine on two rasters with different cell size

DitlevReventlow — Thu, 16 Jul 2015 08:53:37 GMT

Thank you Dan. The thing is that I have already done it and there is no time to redo it, even though this would obvious be the best. I know it should have been the bigger raster that should have been resampled to the cell size of the smaller raster.

Therefore, I just need to be aware of the consequences of the way that I have used combine (not using the "resample" tool, but having "combine" do the resampling process using the bigger cell size as output raster").

The importance is what combine has done in this process. Is there really no way to figure this out?

Re: Combine on two rasters with different cell size

DanPatterson_Retired — Thu, 16 Jul 2015 10:12:02 GMT

The consequences of aggregating to a coarser scale/cell size, is the loss of information.

The consequences of producing a smaller cell size by cell division is an increase in the count per class, but not as a percentage of the total.

The consequences of using any 'filter'...mean, majority, minority etc etc ... well it depends...

at worst, representation 'issues/errors' ie 3x3 window ==>

average of 100,1,1,2,2,1,1,1,1 Hmmmmm?

majority of ............................ not bad for this one but can't say it is good for all

etc

The consequences of interpolation, depend upon the nature of the data, its values and spatial arrangement and the interpolation method used. There are guidelines...but people tend to do ... "guidelines = fact" so I won't give any and recommend that you examine the results of the methods and make a decision based upon your results.

Re: Combine on two rasters with different cell size

DitlevReventlow — Thu, 16 Jul 2015 10:33:33 GMT

Thank you for this Dan. I believe I understand the general consequences of increasing or decreasing cell size through resampling.

What I am unsure about is which interpolation/resampling method that "combine" uses? I have tested all the resampling method, except majority, and none of them gave the same results as combine. And this is even though I have adjusted the processing extent and the snap raster that that they fit.

The reason why I havent tested majority is because there is not drawn any raster when I resample with majority. So I cannot check if the pixel values are the same as in the combined dataset.

So what I am unsure about about is why the majority resampling method does not work, and if there is anywhere that I can check how combine resamples?

I know it would be better to do it all over again (resample and then combine) but I dont have time to do this. I just need to know what happened (how information exactly was lost).

Re: Combine on two rasters with different cell size

DitlevReventlow — Thu, 16 Jul 2015 10:34:50 GMT

And if it is possible that combine uses another resampling/interpolation method that is not part of the 4 options in the "resample" tool.

Re: Combine on two rasters with different cell size

GabrielUpchurch1 — Thu, 16 Jul 2015 19:59:11 GMT

It appears to be using Nearest Neighbor resampling. I did some testing with data I specifically fabricated for this purpose and that is what my results indicate. It is grabbing the cell value from the input rasters that is closest to the cell center of the output raster.

Re: Combine on two rasters with different cell size

DitlevReventlow — Thu, 16 Jul 2015 21:22:58 GMT

I thought it was very strange that I didn't get the right result when i did nearest neightbour resampling but now I have the solution to the mystery.

It does indeed use nearest, but not in the same way as "resample". Even though the processing extent is entirely the same and I have used snap raster it doesn't perform the operation in the same way in the two case.

Combine selects the pixel above the pixel selected "resample" with nearest neighbour. Furthermore, while resample gives you the precise number, combine rounds down even though it is not correct to do so. This means if the nearest pixel is 17,9, then the new value will be 17.

Very strange indeed. Thank you very much for your help Gabriel in the clarification of this!

Re: Combine on two rasters with different cell size

DanPatterson_Retired — Thu, 16 Jul 2015 21:53:57 GMT

it probably does truncation rather than rounding since the output of combine has to be an integer AND sequential...rounding could skip classes