Image Service File Size is Very Large

[JeffMoen]

We recently created an image service from a mosaic dataset of our 2017 ortho images.  We have cached it at all scales (125,000 to 125).  At the 125,000 scale there are 4 tiles at a size of 0.15 MB.  At the 125 scale there are 1,924,852 tiles as a size of 51.96 GB.  We also have has an image service created the same way of our 2015 orthos, however the file size of the 125 scale is only 8.26 GB.  the 2017 source images are jpegs, whereas the 2015 source images are ECW, but the file sizes and numbers of the source images are almost identical between the years.

NOTE: 2015 tile format is JPEG (96 DPI) and 2017 tile format are PNG24 (96 DPI)

We are trying to figure out why the 2017 ortho image service is so large compared to the 2015's, and how we can make it smaller.  Please let me know if there is any other information required to diagnose this!

Thanks,

[SteveCole]

You don't mention this but one possible explanation would be if the orthos are of different resolution, specifically, if your 2017s are a higher resolution than your 2015s.

[JeffMoen]

Thanks for the note, I've added more info to my original post. 

"NOTE: 2015 tile format is JPEG (96 DPI) and 2017 tile format are PNG24 (96 DPI)"

[SteveCole]

I was more referring to whether they were 6 inch resolution vs 1 foot vs 1 meter, etc. More pixels = more disk space.

[JeffMoen]

They are both 10 cm resolution ortho photos

[CodyBenkelman · Accepted solution]

This could be your answer.  The PNG format tiles are not compressed.  We recommend using "mixed JPGPNG" format, which uses JPG for all tiles *except* those that have pixels labeled as transparent.  Using any compression method (JPG, ECW, MrSID, others), if a pixel is 0, the compression may change that to a 1 or a 2 (still very dark black).  If 0 is supposed to be treated as transparent, the compressed nonzero values will no longer treated as NoData in ArcGIS (a.k.a. transparent).  For those (typically few) tiles with NoData values, they'll be written as PNG, but the majority will be JPG.

Try reprocessing a SUBSET (not the entire dataset!) and use format = JPGPNG.

If you have a lot of NoData pixels within the interior of your dataset, this may require further discussion.

[AdrianWelsh]

Is it possible to make your files into MrSID files for greater compression?