Yes, you could use Arc’s model builder to automate the calculation of USLE, and, realistically, you could probably rebuild the entire SDR model using Arc’s various geoprocessing toolboxes. In fact, that’s exactly how InVEST used to be implemented: as an Arc toolbox. We quickly ran into issues, though, and InVEST as it stands is far faster, easier for us to support and maintain, works on Windows, Mac and Linux and can operate on very large datasets, even on memory-constrained systems.
So yes, you could produce a similar sed_export.tif file using only ArcGIS if you just connect the appropriate geoprocessing functions.
The decision to switch from D8 to D-Infinity was so that we could more accurately model water flow across the landscape. D8 assumes that all water flows to the most-downhill pixel, which is then further biased by the diagonal pixels being ‘farther’ away (the distance between the center of a neighbor pixel on a diagonal and the source pixel is longer than the pixel width), which tends towards these long diagonal lines across the landscape. D-Infinity allows for water to flow into multiple pixels at once, and so generates more realistic maps of water flow, but the specification of D-Infinity is actually missing a number of edge cases that makes implementation quite challenging. The routing model used in InVEST 3.7.0 is actually a third one (“Multiple Flow Direction” or “MFD”) that models fractional flow between pixels. You’ll see in your outputs that streams now appear more realistic, and in some maps you’ll even see water flowing around islands that are mid-stream in a river. All of these have been implemented as open-source python packages so that we can use them without overbearing licensing constraints on multiple operating systems and on very large datasets, and the D8 and MFD implementations are available in the python package pygeoprocessing if you’d like to use them outside of InVEST.