Historical patterns of debris flows have been reconstructed at the town of Forest Falls in the San Bernardino Mountains using a variety of field methods (mapping flow events after occurrence, dendrochronology evidence, soil chronosequences). Large flow events occur when summer thunderstorms produce brief high-intensity rainfall to mobilize debris; however, the geomorphic system exhibits properties of non-linear response rather than being a single-event precipitation-driven process. Previous studies contrasted the relative water content of flows generated by varying-intensity summer thunderstorms to model factors controlling flow velocity and pathway of deposition. We hypothesize that sediment discharge in this geomorphic system exhibits multiple sources of complexity and present evidence of (1) thresholds of sediment delivery from sources at the higher reaches of bedrock canyons, (2) storage effects in sediment transport down the bedrock canyons, and (3) feedbacks in deposition, remobilization, and transport of sediment across the alluvial fan in dynamic channel filling, cutting, and avulsion processes. An example of the first component occurred in March 2017, when snowmelt generated a rapid translational landslide and debris avalanche of about 80,000 m3; this sediment was deposited in the bedrock canyon but moved no farther down gradient. The second component was observed when accumulation of meta-stable sediments in the bedrock canyon remained in place until fluvial erosion and subsequent debris flow provided dynamic instability to remobilize the mass downstream. The third component occurred on the alluvial fan below the bedrock canyon, where low-water-content debris flows deposited sediments that filled the active channel, raising the channel grade level to levee elevation, allowing for subsequent spread of non-channelized flows onto the fan surface and scouring new channel pathways down fan. A conceptual model of spatial and temporal complexities in this debris-flow system is proposed to guide future study for improved risk prediction.