When analyzing time series data such as streamflow and associated solutes or sediments, researchers and managers are often interested in studying water quality constituent behavior during isolated storm events when physical processes are most dynamic and more easily inferred. While large scale data analysis at the event scale offers the potential to enhance our understanding of the linkages between event responses and watershed traits, the event-delineation and C-Q response classification pose challenges and typically require manual or semi-automated approaches given the high degree of variation across both space and time.
This presentation focuses on progress to date in event-based research (both event delineation and C-Q pattern classification). We will highlight a new web-based tool, Hydrological Event Detection and Analysis (HEDA), for making event delineation less subjective and more accessible. Using a dataset of over 1,250 storm events from the Lake Champlain Basin, we developed a deep learning convolutional neural network capable of classifying event C-Q patterns into pre-defined hysteresis categories. The tool was able to efficiently classify events using various pattern categorization schemes and represents an advancement over manual visual classification. A frequency analysis of C-Q patterns across the nine study watersheds suggested linkages between event responses and watershed characteristics. We also identified visual trends in C-Q patterns by creating C-Q pattern heat maps as a method to identify possible shifts in watershed function at different temporal scales (seasonal to annual).