Reproducing extremes in continuous stochastic precipitation series

Weather generators are widely used in impact and risk assessment studies to produce long synthetic series of meteorological variables that reproduce current or future climate statistics and natural variability. Most stochastic weather generators are trained to well reproduce the bulk of the precipitation distribution, but they often fail to adequately represent extremes, leading to poor performance in flood hazard and hydrological risk applications. This limitation becomes particularly critical under climate change, as projected impacts on precipitation are expected to manifest differently for ordinary and extreme precipitation values. Here, we address this issue by integrating parametric Weibull tails estimated using the Simplified Metastatistical Extreme Value (SMEV) approach in ordinary weather generator series using a quantile mapping.

The methodology is tested using the AWE-GEN (Advanced WEather GENerator) model applied to a mountainous case study in Friuli Venezia Giulia (north-eastern Italy), characterized by a mean annual precipitation of ~1650 mm.  The AWE-GEN implements the Neyman-Scott Rectangular Pulse (NSRP) model to reproduce the precipitation process. We generate 500 years of synthetic precipitation at 1 hour resolution for the current climate, and for the horizons 2050 and 2100 under RCP 4.5 and RCP 8.5 scenarios. To this end, we use EURO-CORDEX projections and the Clima Nord-Est platform to estimate the factors of change. Specifically, two different approaches are compared: a stochastic downscaling method implemented in AWE-GEN, which uses the EURO-CORDEX projections to assess the NSRP parameters for the future, and a simplified method that requires direct modification of the NSRP model parameters based on the expected factors of change. The parameters of the Weibull distribution for the future were obtained from transient simulations from a convection-permitting model (Lompi et al., 2025).  The adopted downscaling methods led to significant changes in mean annual precipitation, mean annual number of events and mean intensity per event.

This research received funding from European Union NextGenerationEU – National Recovery and Resilience Plan (PNRR), Mission 4, Component 2, Investment 1.1 -PRIN 2022 – 2022ZC2522 - CUP G53D23001400006.