I don't know how much this might help, since relying on models means they fit and are validated to the circumstances.  If true, perhaps this may be helpful. I would suggest modeling at the catchment or subwatershed scale, and aggregating into larger hydrologic units.  I note that we may have some differences in how we use terms, as catchment to me is a very small hydrologic unit, combining as size increases into tributaries, subwatersheds, watersheds, subbasins, and basins.  If you can aggregate many scales of smaller hydrologic units into larger ones, to eventually produce a composite flow of the river basin, then the process in reverse should disaggregate.  Since most do not think in reverse, this disaggregation process is not easy.  Unfortunately if the model as run at the river basin, there is extreme difficulty to reversing, since the assumptions made, model code, etc. are already embedded.  

Combining this in lowlands with increased potential for wetlands, extended riparian areas, braided or meandering channels, deeper substrates with the potential for higher water tables, hyporheic flow or groundwater inputs just adds to complexity.  I would assume that aggregating downstream is difficult enough, but disaggregating requires thinking in reverse into smaller and smaller hydrologic units.  My assumption would be that the better one knows how to aggregate from small to large, the better one would be able to disaggregate from large to small.  But I do not know this to be true.     

Thanks @Hasen

From what you write I am guessing you are interested in subsurface flow and direct surface runoff into the river channel. In either case, the traditional lumped rainfall-runoff models will not help you and I may go as far to say that there is no chance for you to disagregate the results of such a model (it is like trying to identify concrete bricks from a masive concrete wall).

That being said, you may have to change your modelling strategy. One possibility would be a fully distributed model able to describe both surface and subsurface flow including the so called exfiltration (see for example GEOTOP, TOPMODEL, GSSHA). This would probably be the most direct way.

Another way may be soft coupling or linking of submodells, for example you may model the soil profile next to a river with a groundwater modell, where one part of its boundary conditions are defined by the river channel. In this case you would have to do a stepwise modelling, where results from the groundwater model are then used as inflow/outflow from the river channel, thus defining a new boundary condition for the next groundwater modelling step (this means that the approach is/would be labor intensive).

Finally you have the possibility to do hard coupling of models, but that would require a rather intensiv programming effort. Hope it helps.