Researcher(s)
- Hanmin Yang, Geological Sciences, University of Delaware
Faculty Mentor(s)
- Yao Hu, Department of Geography and Spatial Sciences, University of Delaware
Abstract
Coastal communities in the United States face increasing risks from extreme water levels, which threaten infrastructure, residential areas, and emergency-response systems. Sea-level rise and changing storm climatology may alter the magnitude and variability of coastal extremes, making the stationarity assumption of extreme-value models unrealistic. Yet coastal stations do not respond uniformly, and nearby locations may exhibit different skew-surge behavior.
This study develops a deep temporal clustering and hierarchical Bayesian framework to identify groups of U.S. tide-gauge stations with similar skew-surge characteristics and quantify their non-stationary coastal hazards. Using 71 years of annual maximum skew-surge observations, deep temporal clustering model extracts key temporal characteristics from each station’s long-term skew-surge time. Stations will be grouped according to similarities in extreme-event magnitude, variability, and temporal evolution. This regionalization may identify noncontiguous stations that exhibit comparable extreme-water-level.
The resulting clusters will be incorporated into a hierarchical non-stationary generalized extreme value model. Within each cluster, stations will share information while retaining station-specific characteristics. The location, scale, and shape parameters will vary over time, allowing the model to identify non-stationary patterns shared across stations within each cluster, while also capturing station-specific changes.
By combining deep temporal clustering with hierarchical extreme-value analysis, this framework provides a data-driven approach for regionalizing U.S. coastal extremes based on observed behavior. The results can improve understanding of how coastal hazards differ across stations and support infrastructure design and risk assessment.