John
this makes sense.
I'd encourage you to try doing all of this in ArcGIS Pro. (If you don't have Pro now, you can get a trial copy)
It may be possible that a change in your workflow could output 16 bit integers from Pix4D, but if you have 70 surveys you don't want to have to reprocess all of those through Pix4D.
You should not need to convert to 16bit. If you creating a mosaic dataset of all the imagery, you can treat it as one logical collection. The mosaic dataset can also combine any desired bands into a multispectral view (e.g. 4,1,2 for color IR, 4 & 1 for NDVI, 5 & 1 for red-edge-NDVI, etc.), and this is done instantly via a function, not a process you need to run and create more output files.
For the classification, you or colleagues can use advanced classification tools and wizard in the Image Analyst Extension or Spatial Analyst. (also available as a free trial for ArcGIS Pro). This can use any bit depth, or you can apply an arithmetic function prior to classification to convert (on the fly) to 16 bit.
If you really need to proceed with converting to 16bit, then you can use OptimizeRasters.
Download from http://esriurl.com/OptimizeRasters and use following command
c:\Image_Mgmt_Workflows\OptimizeRasters\OptimizeRasters.py -config=Imagery_to_TIF_float2int.xml -input=c:\tmp\Q\inputdirectory -output=c:\tmp\Q\outputdirectory
(See http://s3.amazonaws.com/ImageManagementWorkflowsTeam/OptimizeRasters/Imagery_to_TIF_float2int.xml )
.png "JieZhang"){kind=avatar}
