Evaluation and bias correction of simulated sub-daily rainfall extremes by regional climate models

Sub-daily precipitation extremes can have a huge impact on society as they cause hazards such as flooding, erosion and landslides. For example, the July floods in Germany, Belgium and nearby countries, were one of the costliest events in Europe, with insured losses up to USD 13 billion. Climate change is expected to intensify precipitation extremes as atmospheric water content increases by 6-7% per degree of warming, underscoring the need to predict future hydrological hazards. Regional Climate Models (RCMs) typically run at a spatial resolution of 12-25 km, but they insufficiently describe the small-scale features of observed sub-daily precipitation extremes. The past decade, convection-permitting RCMs were developed which run at high resolution (1-4 km), and explicitly resolve deep convection. Confidence in future projections requires that RCMs adequately simulate the statistical features of observed sub-daily extreme precipitation and also represent the physical processes associated with convective events. We propose a diagnostic framework for simulated 1h-24h rainfall extremes that summarizes the overall RCM performance. This includes the following metrics: the seasonal/diurnal cycle, temperature and humidity dependency, temporal scaling, and spatiotemporal clustering. A substantial part of the work is devoted to the statistical modelling of the metrics with Extreme Value Theory (EVT). We illustrate the evaluation tool with convection-permitting RCM simulations over Belgium against high-frequency observations and assess the benefit of the convection-permitting RCMs with respect to coarser scales. Finally, we give some guidelines for bias correction of simulated precipitation extremes.Sub-daily precipitation extremes can have a huge impact on society as they cause hazards such as flooding, erosion and landslides. For example, the July floods in Germany, Belgium and nearby countries, were one of the costliest events in Europe, with insured losses up to USD 13 billion. Climate change is expected to intensify precipitation extremes as atmospheric water content increases by 6-7% per degree of warming, underscoring the need to predict future hydrological hazards. Regional Climate Models (RCMs) typically run at a spatial resolution of 12-25 km, but they insufficiently describe the small-scale features of observed sub-daily precipitation extremes. The past decade, convection-permitting RCMs were developed which run at high resolution (1-4 km), and explicitly resolve deep convection. Confidence in future projections requires that RCMs adequately simulate the statistical features of observed sub-daily extreme precipitation and also represent the physical processes associated with convective events. We propose a diagnostic framework for simulated 1h-24h rainfall extremes that summarizes the overall RCM performance. This includes the following metrics: the seasonal/diurnal cycle, temperature and humidity dependency, temporal scaling, and spatiotemporal clustering. A substantial part of the work is devoted to the statistical modelling of the metrics with Extreme Value Theory (EVT). We illustrate the evaluation tool with convection-permitting RCM simulations over Belgium against high-frequency observations and assess the benefit of the convection-permitting RCMs with respect to coarser scales. Finally, we give some guidelines for bias correction of simulated precipitation extremes.C