Integrating Nonstationarity and Uncertainty For Quantifying Flood Protection Reliability

This presentation outlines techniques, applications, and implications for integrating nonstationary flood frequency analysis with uncertainty in flood peaks, geomorphic adjustment, and land use to quantify the distribution of flood protection reliability over a planning horizon. Deterministic model estimates of reliability mask inherent uncertainties, often assume stationarity, and potentially underestimate communicated failure likelihood. For instance, flood magnitude-frequency relationships are uncertain and can change through time. The flow capacity of rivers and floodplains is also uncertain and may change through time due to aggradation/degradation, land use change, and others. Quantifying the distribution of flood protection reliability reveals the range of uncertainty and enables decision making based on contextually appropriate level of risk. Numerous methods have emerged for conducting nonstationary flood frequency analysis based on trends in flood peaks. Monte-Carlo simulations of flood hydraulics with HEC-RAS provide a computationally efficient method to quantify uncertainty in channel and floodplain flow capacity. The presented approach incorporates channel and floodplain capacity uncertainty into a reliability estimate by numerically integrating the marginal probability distributions of flood occurrences and flow capacity within a nonstationary framework such that the mean and standard deviation of the annual peak flood changes with time according to the identified trend. This computational procedure can be implemented in a bootstrap scheme to quantify the distribution of reliability over a planning or to estimate a design flood magnitude based on a desired level of confidence in a prescribed reliability. The presented approach is applied at a levee along the Mississippi River near St. Louis, MO and at a small urban creek in Charlotte, NC where probabilistic floodplain maps were developed that depict the distribution of reliability. Accounting for uncertainty and nonstationarity can substantially reduce flood protection reliability and indicates that typical, deterministic estimates communicate an optimistic valuation of flood risk. Results show a marked difference in design flood magnitudes depending on the desired level of confidence in reliability and whether nonstationarity and uncertainty are considered. The approach presented here can serve as tool for flood risk management and design, provide a template for nature-based solutions to reduce flood risk, and aid in the design of resilient infrastructure. The results highlight the implications of considering uncertainty and nonstationarity on estimates of current and future flood risk.