I thought I was subscribed to this thread, but wasn't.
The DEM is a bit under 3.5e8 pixels. Most of this is a single drainage basin, and near the outflow, the flow accumulation model would add small accumulations from small watersheds to the large accumulations from the mainstem. A standard 32-bit single precision float has approximately 7(.2) significant digits of precision, so when flow accumulation from the mainstem reaches somewhere around 1e6 or 1e7, flow accumulation will start to lose some accumulation and will misrepresent which pixels have greater accumulation than which other pixels. By way of conceptual example, in the land of single-precision floats, 12345678 + 1 = 12345678, and in some cases 12345678 = 12345677. I wish I had an actual example, but all the numerical software I use only does double-precision.
In short, the results would be wrong, and it would be distracting to have to consider and then explain why software inadequacies are hopefully not affecting any conclusions. This would all be moot if the flow accumulation tool actually used double-precision, but the documentation just says FLOAT and floating point type, which in my experience has usually meant 32-bit single precision.
Oliver
