Simulation of Record-Breaking Precipitation Events Using an Advanced Stochastic Weather Generator

Over the past few decades, increased record-breaking precipitation events have occurred in many places worldwide, leading to devastating flood disasters. Conventional design criteria for hydraulic infrastructure and flood mitigation projects are generally dependent on the analysis of historically observed data to inform projections of future conditions through fitting a probability distribution. Beyond the conventional stationarity assumption, it is also assumed that the past observed extreme data can approximate well the entire statistical distribution of future events of extreme precipitation. However, conventional methods solely relying on an extreme value analysis have been shown to fail to capture record-breaking precipitation extremes, potentially underestimating the risk of failure of hydraulic structures and flood prevention measures. This study leverages on the capability of a stochastic weather generator (AWE-GEN) to simulate record-breaking precipitation events at a point-scale by reproducing an ensemble of hourly synthetic precipitation time series that accounts for the intrinsic variability of the rainfall process. Compared with conventional extreme value analysis methodologies, the approach is capable of reproducing internal climate variability well and often reproduces extreme values of precipitation, which have not been recorded in the data yet. This study showcases the relevance of stochastic rainfall generators for estimating precipitation extremes for hydrological design under an uncertain climate.