I would say that in did SWAT is a good candidate, but not necesarily the best and certainly not the only one. Look at the following publication for an overview of about 50 (yes, 50 such modells).
Being that you are interested on looking into the dynamics (you mention: e.g. specific soil surface characteristics evolve in terms of shape, size, location in the watershed) I would suggest you go for a more physically based approach. Note that the EROSION module of SWAT with the USLE equation and its derivates MUSLE, RUSLE is purely empirical.
You may also want to look at so called "landscape evolution models" like CHILD and CAESAR. Such models are usually developed for very large time scales in mind (thousands of years) but the concept is precisely that of a dynamic lanscape as the one you are interested in. In the case of CAESAR it is supossed to be applicable at shorter time scales (days - years), so a look into them may be worth effort.