A Hybrid Framework for Rapidly Locating Transition Zones: A Comparison of Event- and Response-Based Return Water Levels in the Suwannee River FL

Flood risk assessments commonly use event-based approaches to reduce the number of scenarios required to be run through computationally intensive physical process models. Often the return period of the response variable (e.g., a fluvial water level or overtopping discharge) generated by an event (e.g., upstream/downstream water level or set of sea state variables) does not match that of the event itself; a limitation of event-based approaches which can lead to the misspecification of flood risk. We present a transferable hybrid statistical-hydraulic modeling framework for rapidly locating transition zones; river reaches where extreme water levels are driven by both upstream riverine discharge and downstream sea level. Instead of an event-based approach the framework utilizes a surrogate model to reduce computational expense of the hydraulic model. The surrogate-based approach allows the empirical estimation of response-based along-river return levels from a large number of plausible discharge-coastal still water level events simulated from the statistical model. We assess the robustness of the event-based approach by comparing the associated return levels with the response-based return levels. The framework is applied to the Suwannee River in Florida (United States). Three surrogate models are evaluated, highlighting the enhanced ability of non-linear models to accurately capture discharge-sea level interactions along the river. The along-river return levels of the most-likely design event are found to lie within the range of variability of the response-based return levels for most of the transition zone.

Automated summary

This study presents a transferable statistical-hydraulic modeling framework for rapidly locating transition zones where water levels are influenced by both upstream river discharge and downstream sea level. The framework reduces computational cost through surrogate modeling and evaluates the robustness of the event-based approach by comparing it with response-based return levels.