The behaviour of the Split Raster function is rather unpredictable and produces partial or no output depending on the output raster data format chosen. Some tiles would complete successfully while others tiles would either contain NODATA or be skipped altogether. This would happen when producing stand-alone GRID's, rasters within file geodatabases, and also TIFF's (curiously, the pattern of which tiles were successful and which weren't was not consistent between formats). The only format that was successful in my trials was IMAGINE IMAGE (.IMG) which produced data for all the tiles successfully. Further testing would be required to determine if this is always the case.
I searched all the forums and found one entry by a user experiencing similar problems, but the solution to the issue was never pursued as the user found a manual 'work around' which doesn't work for me as I have to do this repeatedly for large areas in Ontario. Forum reference: http://gis.stackexchange.com/questions/211630/missing-output-from-arcgis-split-raster-tool
Illustration below was the result using the stand-alone GRID format but there were also instances where no tiles were created at all in other tests.
The following are the DEM characteristics and the splitting criteria I used:
DEM Projection: NAD_1983_CSRS_UTM_Zone_17N
Data Type: Float 32-Bit
Unsplit DEM Width Size (columns) = 11163
Unsplit DEM Height Size (rows) = 11519
Cell Width Size ( x ) = 2
Cell Height Size ( y ) = 2
Linear Unit Name = Meter
Split DEM Width Size (columns) = 4000
Split DEM Height Size (rows) = 4000
1 PIXEL overlap between tiles.
NEAREST resampling option
I also encountered this problem using a DEM with a 30m cell size, NAD_1983_Ontario_MNR_Lambert projection, and on the following platforms.
I am wondering whether later versions of ArcGIS have fixed this problem?
Thanks in advance.