Difference between "Project raster" and "Resample" to resample raster

3721
4
06-17-2015 11:58 PM
Highlighted
Occasional Contributor II

Hey everyone.

I was working with a raster layer in WGS_1984 and wanted to convert this to to Sinusoidal (sphere) and resample it to half the original cell size.

I thought It was really smart when I discovered that you could do this in one go with "Project raster" tool. But then I could see that pixels were being somewhat distorted using this method.

I then tried to do it step by step - meaning first converted to sinusoidal with "project raster" and then using "resample" to change the cell size. This result from this look interely different.

I have attached a few pictures of the scenarios and would be very grateful if someone knew what caused this problem.

Thank you very much.

Kind regards

Ditlev Reventlow

Tags (2)
Reply
0 Kudos
4 Replies
Highlighted
MVP Frequent Contributor

Well both tools are "resampling" the input pixels to map them to the output. Looks like you used nearest because your input looks like a binary image.

Results will change depending on the input/output projection and the cell size and alignment.

There is no "wrong way" really here. Did you try the anchor point options to align the outputs to an exact point?

Highlighted
Esri Esteemed Contributor

It would have been helpful if you would have included the different parameters specified in the project raster and resample tools. Judging from the effect and you data (0, 1?) you are using the nearest resampling technique (which is OK for this type of data). The original data is visualized using a on-the-fly projection (skewed pixels). I also see that the area has been moved south about 6 pixels, did you move the area, or is this what happened after projection (that would be something to check). Is there a specific transformation you apply?

Another question: what does the data represent? The reason I ask is that if the dataset is not too big and it originally represents points or areas, you could consider converting back to vector format, project and then rasterize. Or even much better, if you have access to the original vector input data you should use this instead (project and then convert to raster)

One of the differences you could see, is when you first resample and then project. This will create a different result. Projecting the raster first, yields "comparable" single pixels. If you resample afterwards (factor 0.5) there will be 4 output pixels for each input pixel, but graphically you don't see the difference.

Doing this in a single step applies a different order. It is projected and resampled in memory and then the output pixels are assigned based on the center of input pixels.

Highlighted
Occasional Contributor II

Thank you for you help.

Yes I used nearest to resample.

Neil,

I did not use the anchor point option. Is this something you specify in "environments" ?

Xander,

I am extracting Evapotranspiration data from pixels that havent changed their land use. The movement of the data was caused by changing the projection from WGS_1984 to sinusoidal. There is no specific transformation to use for this transformation. The program automatically select "new geographic transformation".

Great, makes sense that it uses the center of the input pixels as a reference. And I guess that I shouldn't do it then.

But it's quite annnoying, I had done a lot of analysis with the map projected and resampled in one go.

Reply
0 Kudos
Highlighted
Regular Contributor

I received some advice on here a while ago that, with ESRI tools, it is always best to perform analysis using many small steps rather than fewer large steps.