Identification of patterns of hysteresis in suspended sediment-discharge relationships to infer watershed sediment dynamics

Studying the hysteretic relationships embedded in high-frequency suspended sediment concentration and river discharge data over individual storm events provides insight into the drivers and sources of riverine sediment during events. However, existing research remains limited to analyses using simple visual classifications (linear, clockwise, counter-clockwise, and figure-eight patterns) or the collapse of these patterns to a hysteretic index. In this study, we leverage three-years of high-frequency suspended sediment, discharge, and meteorological monitoring from within the Mad River watershed to demonstrate a new machine learning based approach to classifying storm events based on the type of hysteresis observed. The main stem of the Mad River and five of its tributaries were monitored between 2013 and 2015 providing 600 unique events to analyze. Fourteen different types of hysteresis were identified within the data set. Events were classified automatically by training a restricted Boltzmann machine (RBM), a type of artificial neural network, on images of the suspended sediment-discharge plots. The probabilistic RBM classification network predicted the correct or next most similar class 71% of the time.

The expanded classification system allowed for new insight into drivers of hysteresis types including spatial scale, antecedent conditions, hydrology and rainfall. Additionally, differences in the type and frequency of hysteresis type were observed between sites and between seasons. This provided insight into differences in timing and source proximity of sediment loading between subwatersheds. With increased availability of high-frequency suspended sediment data, the hysteretic classification approach presented here can be used to inform watershed management efforts to identify sediment sources and reduce fine sediment export.