Probabilistic projections of winter floods considering cumulative effect of air temperature changes

Ice-induced winter flooding, intensified by sustained low temperatures, holds the potential for severe natural disasters, but is seldom explored probabilistically considering warming climate impacts. This study established both marginal and copula-based joint probability distributions of the upstream (QH) and downstream (QL) ice-induced floods in the Lower Yellow River, a hanging river above the ground, under four parametric scenarios (constant, time as covariates, mean air temperature as covariates, and accumulated negative air temperature as covariates), to compare historical and design flood regimes using six inference methods (UNI, OR, AND, KEN, SKEN, and COND) under air temperature changes. The results show that the Lognormal and Weibull marginal distribution models with accumulated negative air temperature as covariate parameters were optimal for QH and QL, respectively and the positive dependence between QH and QL was best described by the Gumbel-Hougaard copula. Impacts of increasing air temperature on flood downtrends and yearly change-points (1990 for QH and 1985 for QL) reduced both historical QH-QL flood magnitude combinations and projected return periods, thus denoting declining flood severities over time. Due to such flood downtrends, the most probable composition (MPC) values of 100-year design floods varied from the highest (1656 m3/s for QH and 1645 m3/s for QL using the OR method) to the lowest (624 m3/s for QH and 342m3/s for QL using the SKEN method). The average decreasing rates of MPC values before and after the discerned flood change-points were 17.4% for QH and 39.6% for QL. When conditioned on the occurrence of upstream QH having flood magnitudes less than 100-year design floods, large floods downstream exceeding a 50-year return period were inferred as improbable. This study can provide a paradigm of flood projections to meet diverse flood control objectives under changing climate.