2026 CSDMS meeting-008

Edit registration

Pay registration

(if you haven't already)

Log in (or create account for non-CSDMS members)
Forgot username? Search or email:CSDMSweb@colorado.edu

Browse  abstracts

<<

>>

Back to meeting

Quantifying the Role of Floods on Chesapeake Bay Biogeochemistry and Sediment Transport Using K-Means, DBSCAN and Spatially Organized Maps

Julia Moriarty, CU Boulder Boulder Colorado, United States. julia.moriarty@colorado.edu

The impact of floods on biogeochemistry and sediment transport varies within individual estuaries and among different systems. This variability, as well as limitations in observational approaches, motivate the use of high-resolution numerical models to better understand how estuaries respond to extreme events and other perturbations. However, identifying and quantifying spatiotemporal patterns in these data are often challenging because of the immense amount of information that these models produce.

This study will therefore leverage unsupervised, explainable machine learning to identify and quantify complex spatiotemporal patterns caused by floods in the Chesapeake Bay using a numerical model. Specifically, K-Means, DBSCAN and Spatially Organized Maps will be used to analyze results from a previously run implementation of the Regional Ocean Modeling System (ROMS) that accounts for hydrodynamic, sediment transport and biogeochemical processes.

Preliminary results indicate that classification of the Bay into spatial regions varies among salinity, phytoplankton concentration, and suspended sediment concentrations clusters. Clustering based on time-series of surface salinity is strongly related to mean values of salinity. In contrast, time-series of surface phytoplankton concentrations became clustered based on event responses and seasonal trends, as well as mean values. Finally, clustering of suspended sediment concentration time-series was based on locations’ responses to the storm, but the results were skewed by outliers, motivating future work to consider the DBSCAN approach instead of K-Means. Ongoing work also includes identifying different phases of events using self-organized maps, considering additional events, comparing results from the entire Bay to results from smaller regions such as the Patapsco River Estuary, and the use of improved machine learning techniques.