B. P. Bledsoe

Recent catastrophic flood events, increasing flood losses, and climate change challenge current reliability estimates defined by the probability that flood levels will not exceed protection measures over a planning horizon. These estimates depict an expected reliability that mask uncertainty in streamflow and the capacity of river channels and floodplains. We described reliability as a random variable whose distribution depends on uncertainty and nonstationarity in annual maximum flood (AMF) distributions and flow capacity uncertainty. Numerical experiments quantified the impacts of nonstationarity and variance in AMFs and flow capacity uncertainty on estimates of flood protection reliability, thereby providing the first examination of their interacting effects. The distribution of reliability along a regulatory floodplain boundary was quantified through a bootstrap scheme that accounts for nonstationarity and uncertainty in AMFs and flow capacity uncertainty. Results indicated that accounting for uncertainty in flow capacity substantially reduces reliability compared to estimates based solely on flood likelihood; the divergence is greater in the presence of nonstationary AMFs. The distribution of reliability along the regulatory floodplain boundary was spatially heterogeneous due to within‐reach variation in flow capacity uncertainty. Quantifying the distribution of reliability for flood protection measures enables transparent communication and selection of a desired confidence level that is commensurate with a contextually appropriate risk tolerance. Similarly, we show that a desired level of confidence in reliability can be specified to estimate a design flood protection level. These results reveal how the combined impacts of uncertainty and nonstationarity can impact reliability estimates and confidence in those estimates.