A New Machine‐Learning Approach for Classifying Hysteresis in Suspended‐Sediment Discharge Relationships Using High‐Frequency Monitoring Data

Abstract

Studying the hysteretic relationships embedded in high‐frequency suspended‐sediment concentration and river discharge data over 600+ storm events provides insight into the drivers and sources of riverine sediment during storm events. However, the literature to date remains limited to a simple visual classification system (linear, clockwise, counter‐clockwise, and figure‐eight patterns) or the collapse of hysteresis patterns to an index. This study leverages 3 years of suspended‐sediment and discharge data to show proof‐of‐concept for automating the classification and assessment of event sediment dynamics using machine learning. Across all catchment sites, 600+ storm events were captured and classified into 14 hysteresis patterns. Event classification was automated using a restricted Boltzmann machine (RBM), a type of artificial neural network, trained on 2‐D images of the suspended‐sediment discharge (hysteresis) plots. Expansion of the hysteresis patterns to 14 classes allowed for new insight into drivers of the sediment‐discharge event dynamics including spatial scale, antecedent conditions, hydrology, and rainfall. The probabilistic RBM correctly classified hysteresis patterns (to the exact class or next most similar class) 70% of the time. With increased availability of high‐frequency sensor data, this approach can be used to inform watershed management efforts to identify sediment sources and reduce fine sediment export.

Monitoring the riverine pulse: Applying high‐frequency nitrate data to advance integrative understanding of biogeochemical and hydrological processes

“…Recent investigations suggest that the HI is being rapidly adopted by those who study nutrient runoff patterns during high flow (Baker & Showers, ; Blaen et al, ; Dupas et al, ; Fovet et al, ; Vaughan et al, ). Limitations of the HI have been discussed and a pattern recognition method using machine learning was recently pioneered to delineate suspended sediment hysteresis typology during events (Hamshaw, Dewoolkar, Schroth, Wemple, & Rizzo, ). Further advances in NO 3 − hysteresis investigations may necessitate detailed modeling that includes empirical representation of controls on net supply applied to formally test process‐based hypotheses as developed in turbidity studies (Mather & Johnson, 2014).…”

Exploring Deep Neural Networks to Retrieve Rain and Snow in High Latitudes Using Multisensor and Reanalysis Data

“…Recent development of deep learning techniques brings great opportunities for the scientific community of water sciences (Hamshaw et al, ). The utilization of deep learning methods is shown to be useful in resolving some traditionally difficult problems.…”

Impact of an Extreme Storm Event on River Corridor Bank Erosion and Phosphorus Mobilization in a Mountainous Watershed in the Northeastern United States

“…To contextualize our estimates from the six Irene‐affected stream bank sites studied here in terms of watershed‐scale sediment and TP dynamics, we compared our estimates to the following: (a) areal channel‐change measurements conducted over a greater length of the main stem for the same time period (2008–2011) (Jordan, ) and (b) watershed‐scale suspended sediment and TP flux data estimated using data collected during other studies (Hamshaw et al, ; Jordan, ).…”

“…We generated an annual average estimate based on 19 years of data prior to Irene and separately report Medalie's estimate for 2011, which includes the Irene flood event. Second, an estimate of average annual suspended sediment export was derived from data collected in a separate study by Hamshaw et al (2018). TSS load from the Mad River watershed was estimated seasonally (early spring through late autumn) from 2013 to 2016 by surrogate monitoring using a turbidity sensor.…”

“…TSS load from the Mad River watershed was estimated seasonally (early spring through late autumn) from 2013 to 2016 by surrogate monitoring using a turbidity sensor. To scale the seasonal TSS load estimates to the entire year, turbidity-based measurements from Hamshaw et al (2018) were combined with estimates for the period when direct monitoring was not available (i.e., spring melt period) by referencing discharge data from the USGS stream gauge and utilizing existing regression relationships of TSS to discharge (Hamshaw, 2018). Average annual TSS export for the Mad River was then calculated using the cumulative TSS load from the four-year (2013 to 2016) period.…”