Case-Selective Dynamical Downscaling for Efficient Extreme Precipitation Statistics at Convection-Permitting Scale

High-resolution climate simulations with convection-permitting models (CPMs) are essential for studying sub-daily precipitation extremes, but their computational cost severely limits the length, domain size, and ensemble size of continuous simulations. This poses a major challenge for obtaining robust extreme-value statistics at kilometre scale. Here we introduce a case-selective dynamical downscaling (CSDD) framework that enables the reconstruction of extreme precipitation statistics at convection-permitting resolution without requiring long, continuous CPM simulations.

The approach identifies time windows likely to contain extreme rainfall using precipitation from a coarser-resolution driving simulation, and dynamically downscales only these selected periods. Applied to a 30-year regional climate simulation, CSDD reproduces the statistical distribution of 1–6 hour precipitation extremes from a full continuous CPM simulation while requiring only about 10 % of the computational cost. Because individual cases are independent, simulations can be executed fully in parallel, allowing wall-time reductions of several orders of magnitude and facilitating ensemble-based uncertainty quantification.

Our results demonstrate that reliable kilometre-scale extreme precipitation statistics can be obtained without continuous CPM integrations, making CSDD a complementary strategy to traditional regional climate modelling. By alleviating key computational bottlenecks in long-term CPM applications, the framework enables efficient ensemble generation for extreme-precipitation research and opens new opportunities for extreme-event analysis at convection-permitting resolution.