205 - Spatially Varying and Duration Dependent Covariate Model: A Hierarchical Bayesian Framework for Multi-duration Extreme Precipitation Frequency Analysis in Texas

Intensity-Duration-Frequency (IDF) curves are widely used in engineering design, risk assessment, and floodplain management. Although it is expected that climate change will alter the characteristics (i.e. intensity, duration and frequency) of heavy rainfall, available guidance such as Atlas 14 relies heavily on the stationarity assumption. This may underestimate present and future hazards, leading to infrastructure under-design. Prior studies have shown that conditioning the parameters of the statistical distribution on time-varying climate covariates can improve nonstationary estimates. However, this approach increases the number of parameters to be estimated, exacerbating parametric uncertainty, which propagates into highly uncertain projections. Moreover, the scarcity of data, especially for short durations, challenges multi-duration analysis. To address these, we propose a Spatially Varying and Duration Dependent Covariate Model for process-informed precipitation frequency analyses. Specifically, we assume that (1) the robust effects of climate covariates on the probability distribution of extreme rainfall are heterogeneous in space and (2) the distribution parameters are dependent on the durations. We employ a Bayesian hierarchical spatial model, leveraging Gaussian processes and extreme value theory, and apply this model to infer multi-timescale nonstationary rainfall exceedance probabilities in Texas. The proposed approach is highly flexible, naturally allows the use of stations with incomplete observational records across durations, identifies robust temporal trends, estimates IDF curves over a wide range of durations, and transparently quantifies parametric uncertainty.