How reliably can low-flow rates of the Danube to prescribed frequencies be estimated for nuclear power station capacity expansion at Paks?

Abstract

Nuclear power stations must be designed with consideration to extreme events having return periods (T_r) of up to 10⁵ (sometimes even 10⁶ for the largest plants) years. This is a daunting task for occurrences influenced by hydro-meteorological variables, such as floods and stream baseflow rates in the light of climate variability and change. Here we demonstrate the unavoidable large uncertainty encountered in such estimates, further illustrated by the simple task of deriving the T_r value for the latest annual minimum (Q_min) observed at Budapest during the recent Europe-wide drought of 2026 that resulted in record breaking low water levels (but not necessarily in discharge rates) at most gauging stations in Hungary due to a general incision of the river bed into its alluvium. Depending on the homogenization method or absence thereof, the latest Q_min of 662 m³s^-1 observed on October 8, 2026 in Budapest, has an associated T_r of 89 years with a 68% confidence interval of [47; 168] years before homogenization, a T_r of 126 [63; 254] years after homogenization, and 516 [180; 1,477] years when the observed increase in annual minima is also accounted for. As a consequence, any percentile estimate (and especially its inverse task of finding T_r for a given Q_min) based on hypothetical future trajectories of the hydro-meteorological variables must be taken with extreme caution as such estimates must necessarily have a very wide combined confidence interval thus making them highly uncertain.