Stability of Homogeneous Regions Under Record Extension and Its Propagation to Station-Level Rainfall Quantiles in Regional Frequency Analysis

Floods and extreme precipitation events pose growing risks to infrastructure and water resource management, making the reliable estimation of design rainfall quantiles essential. As observational records are extended, the spatial similarity structure underlying regionalization may itself change, with direct implications for regional frequency analysis (RFA), in which reliable quantile estimation hinges on delineating hydrologically homogeneous regions. This study investigates whether the regionalization of extreme rainfall in South Korea remains stable as the observational record is extended, and whether such changes propagate to station-level rainfall quantile estimates. Homogeneous regions were delineated independently for each period using K-means clustering, applied under three dimensionality reduction configurations—principal component analysis (PCA), t-distributed stochastic neighbor embedding (t-SNE), and a baseline without reduction—so that any detected change could be attributed to the data rather than to the analytical procedure. Each was applied to 615 rainfall stations over three cumulative periods (2017, 2021, and 2025), drawing on 49 hydrometeorological variables with silhouette-based cluster selection, and L-moment-based heterogeneity measures. Comparison across periods revealed shifts in cluster boundaries across several inland and coastal areas that were consistent across all three configurations, with estimated rainfall quantiles increasing or decreasing accordingly at individual stations. Between 2017 and 2025, station-level 100-year 24-h rainfall quantiles changed by −10% to +44%, and the mean 200-year 24-h quantile under C#1 increased from 374.6 to 391.5 mm (+4.5%). These results indicate that the delineation of homogeneous regions may change as the observational record is extended, and that these changes propagate to design quantiles. Accordingly, homogeneous regions should not be assumed to be constant over time; they should be re-delineated and re-examined whenever the observational record is updated.