Comparison of Temporal Rainfall Disaggregation Methods for Climate-Change-Adaptive Design Rainfall and Their Applicability to Agricultural Production Infrastructure: Method of Fragments, Multiplicative Cascade, and Copula

Hydrological design of agricultural production infrastructure requires sub-daily design rainfall, yet climate-change projections are mostly daily, making temporal disaggregation indispensable. This study compares three daily-to-hourly methods—the Method of Fragments (MOF), a microcanonical multiplicative cascade, and a Gumbel copula—applied identically at 76 Korean weather stations under Shared Socioeconomic Pathway (SSP) SSP2-4.5 and SSP5-8.5 with a multi-model ensemble (MME) of 18 global climate models. Split-sample tests at eight agricultural stations evaluated the reproduction of observed hourly statistics, storm events, and return-period quantiles. MOF reproduced the observed sub-daily extremes and persistence most closely (out-of-sample intensity–duration–frequency bias −0.5%, Nash–Sutcliffe efficiency 0.95, lag-1 autocorrelation 0.42 versus 0.48 observed), whereas the cascade underestimated (0.12) and the copula overestimated (0.75) the autocorrelation. MOF reproduced 3–6 h storm maxima within 10% but overestimated 1–4 h extremes by 10–23%. For 2031–2070, the cascade inflated the 60 min 200-year quantile to 330–342% of observations, whereas MOF gave moderate increases (6–24% at 12–48 h). These partly implementation-dependent differences identify MOF as best suited to agricultural-infrastructure design with the present data. As new scenarios or methods may change this choice, the transferable contribution is the approach: screening methods out-of-sample against design-relevant extremes before MME application, giving an objective basis for revising design standards.