Hillslope Runoff Attenuation Features (RAFs) are soft-engineered overland flow interception structures utilised in natural flood management, designed to reduce connectivity between fast overland flow pathways and the channel. The performance of distributed networks of these features is poorly understood. Extensive schemes can potentially retain large quantities of runoff storage but there are suggestions that much of their effectiveness can be attributed to desynchronisation of subcatchment flood waves, and that inappropriately-sited measures may increase rather than mitigate flood risk. Fully-distributed hydrodynamic models have been applied in limited studies but introduce computational complexity. The longer run-times of such models also restricts their use for uncertainty estimation or evaluation of the many potential configurations and storm sequences that may influence the timing and magnitude of flood waves. <br><br> We applied a simplified overland flow routing module and representation of RAFs to the headwaters of a large rural catchment in Cumbria, U.K., where the use of an extensive network of such features is proposed as a flood mitigation strategy. The model was run in a Monte Carlo framework over a two-month period of extreme flood events which occurred in late 2015 that caused significant damage in areas downstream. Using the GLUE uncertainty estimation framework, we scored our set of acceptable realisations and these weighted behavioural realisations were rerun with one of three drain-down time or residence time parameters applied across the network of RAFs. <br><br> The study demonstrates that the impacts of schemes comprising widely-distributed ensembles of RAFs can be modelled effectively within such a reduced complexity framework. It shows the importance of effective residence times on antecedent conditions in a sequence of events. We discuss uncertainties and limitations introduced by the simplified representation of the overland flow routing and RAF representation and how it could be verified and improved using experimental evidence. We suggest ways in which features could be grouped more strategically and means by which the synchronisation issue could be addressed.