Evaluating visual classification of suspended sediment – discharge hysteresis via crowd-sourcing and in-stream monitoring

Understanding what controls storm event responses within a watershed (e.g., antecedent conditions, geomorphic condition of river channels) can help determine the sources of fine sediments and sediment-bound nutrients. Addressing the sources of erosion is of paramount importance in addressing eutrophication and harmful algal blooms in receiving waters where excessive fine sediment and nutrient loading is problematic, such as the case with Lake Champlain in northeastern United States. Studying the hysteresis in the suspended sediment–discharge relationship enables storm events to be categorized based on observed hysteresis patterns, and therefore helps characterize dominant sediment dynamics within a river system. This study seeks to (1) determine the reliability of visual interpreting hysteresis patterns and (2) assess the consistency of observed hysteresis patterns along river segments.

Storm events from the Mad River watershed, located in the Lake Champlain Basin, were classified into 14 hysteresis patterns or types. A survey containing 100 hysteresis patterns from these storm events was presented to a group of survey respondents comprised of six domain experts and 22 non-experts. Analysis showed significant variability in classification performance across both individual respondents and among expert and non-expert groups. The survey results allow for refining the 14 hysteresis type categories, quantifying the variability in visual classifications, and developing a benchmark to compare human classification with the performance of machine learning algorithms.

To help validate the use of relatively low-cost, in-stream sensors and the consistency of characterizing suspended sediment dynamics using hysteresis patterns along a short river segment, turbidity sensors were installed at multiple locations along a 3-km segment of the main stem in the Lewis Creek watershed, also located in the Lake Champlain Basin. Turbidity and river stage sensors were placed in different geomorphic settings (sediment regimes) along the river. The storm event sediment responses were analyzed across each monitoring site by determining the turbidity-stage hysteresis pattern. Together, this research helps improve the efficiency of characterizing storm event sediment dynamics and nutrient loading to rivers.