First, check your data to see if it is in the same projection/coordinate system. It is usually best to place all the data in the same projection/coordinate system before processing it. In theory, one could ignore the warning, but it may be that the warning is relevant. Given the uncertainty, reduce that potential by putting all your data into the same system before processing. Data prep on the front end goes a long ways to head off later issues.
001003: Datum conflict between input and output.—Help | ArcGIS Desktop
As to why the watershed may be derived "downstream" of the pour point, to troubleshoot this I would start by looking at the output surface and checking the elevation values. If the elevation values "downstream" of the pour point are the same value as above, that usually indicates a data issue. Check the DEM used as the input to to the Watershed process. Maybe it is too coarse for what is being attempted? For example, the cellsize is some large value like 1 mile when what is needed is a resolution of 10 feet? Or a Fill process run before the Watershed geoprocessing tool caused the area "downstream" of the pour point to be "leveled out". Just examples of what can go awry. You will have to check the data.
Along with this, confirm that the pour point is really in the correct and relevant location. Is it located right at the top of the spilllway of the lake or instead in the lake many feet away? If the latter, that will likely be the issue. Any chance the Snap Pour Point process didn't work out that well? Review the data and see if its location makes sense given other reference information.
Chris Donohue, GISP
