Probabilistic Extreme Rainfall Projections under Climate Change: Extending Regional Extreme Value Analysis with a Climate Model–Informed Bayesian Approach

Abstract

Abstract Climate change challenges traditional civil engineering design approaches that rely on stationarity and use historical data and parametric distributions to quantify uncertainty, e.g., the generalized extreme value (GEV) distributions for extreme rainfall. Key barriers include the application of forward-looking climate information—typically provided by global climate models (GCMs)—to existing, historical observation–based GEV distributions, while handling various sources/forms of climate change uncertainty. This work presents a Bayesian framework by constructing a state space model to describe local extreme rainfall under climate change. This framework links GEV parameters to global warming level (GWL) and integrates endogenous parameters to describe the responses of GWL and subsequently local extreme rainfall from exogenous forcings such as those from greenhouse gases. Implemented via Particle Filter, the parameter estimation procedures involve using climate model simulations to inform parameters and serve as a prior belief, which is then updated with historical observations. Such procedures avoid the use of adjustment factors and associated methodological uncertainty when bridging climate model simulations and observation data. A case study for Seattle was conducted: analyzed across durations and applying equal weighting factors to forcing scenarios (alternative ones can also be used), results suggested an average 15.4% increase in extreme rainfall from 2025 to 2100 (GWL increased from 1.3°C to 3.5°C during the same period). The method also helps quantify the reliability of existing flood-management structures, e.g., a 1-day, 100-year design storm estimated based on average 1850–2025 conditions at the Seattle-Tacoma Airport is projected with an effective exceedance probability of 3.7% per year in 2100. A large uncertainty is associated with such mean projections, and there is notable spatial variability across stations. Given the engineering needs for forward-looking, local extreme rainfall information, this work supplements other ongoing national or large-scale extreme rainfall analyses and provides a useful alternative to project local extreme rainfall.