Projecting identical .tif and file gdb rasters produce different results

[FrancesBiles1]

Hello,

I have 2 identical raster DEMs. The alignment, cell size, extent, and cell values are the same. This can be seen in the attached screenshot of the raster properties (page 1). I also verified this by comparing the pixels on screen. The only property not showing in the attachment is both rasters also have a vertical projection of NAVD88 defined. 

The only difference between the 2 rasters is that one is a .tif file and the other is a file geodatabase raster.

Each raster is then projected (using Project Raster) to ITRF08 (horizontal) and ITRF08 (vertical) using the ~NAD_1983_2011_To_NAVD88_Alaska_GEOID12B_Height + ~ITRF_2008_To_NAD_1983_2011 geographic transformation method.  [note: there is a documented ESRI bug (BUG-000128301) when doing vertical transformations on .tif rasters in ArcGIS Pro. The transformation is incorrect unless the tilde is removed from the 2nd half of the compound transformation string. When I projected the .tif, I removed the 2nd tilde. It is not necessary to remove the tilde in the case of file gdb rasters].

When projecting the 2 rasters all the settings in Project Raster are identical. For both runs a 3rd raster is specified in Project Raster as the snap raster and a cell size raster. Bilinear resampling is used in both. The .tif raster is output as a .tif raster. The file gdb raster is output as a file gdb raster.  

However, when comparing the outputs after projecting, the 2 rasters are slightly different. The pixels are aligned, but the extent and pixel values are different, as can be seen in the raster properties on the 2nd page of the attachment. 

The only difference between the 2 input files is the file format (.tif vs file gdb raster). The Project Raster settings were identical. Why are the 2 output rasters different?? This is creating confusion because subsequent analysis results differ depending on which DEM is used.

I am using ArcGIS Pro 2.6.3.

[DanPatterson]

Pro 2.6 and 2.7 introduced...

How the Cell Size Projection Method environment setting works—ArcGIS Pro | Documentation

which may provide some insight

[FrancesBiles1]

Hi Dan, 

Thanks for the reference. I usually select 'Preserve Resolution', but in this case, where the same Snap Raster is being used for both Project Raster runs, as well as the same raster being specified to use for the Cell Size, wouldn't the Cell Size Projection method be irrelevant?

[DanPatterson]

looks like the one on the left of your 2nd row of images has no, nodata value specified, which may be the case. The differences aren't huge but the maximums etc do indeed differ from the inputs and they aren't even consistent within.  

I wouldn't rule out testing the cell size projection method and check your nodata values.  Identify the areas where the maximum occurs prior to projecting, then again after.  If large areas and not a few speckles are involved then I would be concerned

[FrancesBiles1]

In the properties of a file gdb raster, the NoData entry is blank, however, it is still there. From my understanding, described here, NoData in a file gdb raster is stored as a mask in the dataset rather than using a pixel value. I can display the file gdb raster and specify a color for the NoData value. The NoData cells show up correctly and in exactly the same location as the NoData cells in the .tif raster.

I ran another test setting the cell size projection method = preserve resolution. The result is the same, the projected outputs (extents and cell values) are slightly different (page 3 in the attached). When comparing values on a cell-by-cell basis using the explore tool and clicking across the screen, the values are different throughout the raster. For example, for 2 coincident cells, one is 1.317897 and the other is 1.29102.

For more background, the .tif raster was created from the file gdb raster using 'Export Raster'. So, the tif file is the same as the file gdb raster (extents, cell size, alignment, statistics, cell values are the same), just a different file format. ...Yet, projecting each one yields a different result. Could this be a bug?

Thank you for your insights.

[DanPatterson]

I don't use rasters in a gdb.  esri recommends *.tif.

I would have overtly specified a nodata value in any event so that it was reflected in the raster values.

In any event, I would trust the *.tif before the other based on my readings of

NoData in raster datasets—ArcGIS Pro | Documentation

Mask (Environment setting)—ArcGIS Pro | Documentation

Then there is the whole issue of your projection information perhaps contributing to the differences in z- values

The transformation is incorrect unless the tilde is removed from the 2nd half of the compound transformation string. When I projected the .tif, I removed the 2nd tilde. It is not necessary to remove the tilde in the case of file gdb rasters]

You can try searching Esri Support | ArcGIS Technical Support and file a case if you feel you have covered all your bases, since nothing I have suggested has helped.

Good luck