Interpolation of design rainfall at ungauged locations exploiting the potential of convection-permitting climate models.

Quantifying design rainfall events at varying durations is crucial for assessing flood risk and mitigating losses and damages. Yet, in a changing climate, they are fundamental tool for a reliable design of water related infrastructures, such as flood retention reservoirs, spillways, and urban drainage systems. Usually, design rainfall is quantified where rain gauges are located, and regionalization methods are used to provide estimates in ungauged locations. During the last years, convection-permitting climate models (CPM) are receiving increasing attention because, thanks to their high spatial resolution (~3km) and ability of explicitly resolving atmospheric convection, they allow for better estimating precipitation spatial patterns and extreme rainfall at multiple durations compared to coarser models.

In this work, we combine at-site rain gauge measurements with CPM simulations, within a non-asymptotic statistical framework for the analysis of extreme rainfall. We aim at quantifying the added value of the physics-based information provided by CPM simulations for the estimation of high quantiles of rainfall in ungauged locations.     

The performance of the new regionalization approach is compared with traditional interpolation methods (i.e. interpolation of distribution function parameters) using leave- one-out cross-validation as well as considering different rain gauge densities.

Preliminary results show that the proposed methodology based on CPM simulation provides: i) similar performances compared to traditional gauge-based regionalization methods for high station density scenarios and ii) improved performances for low station density scenarios.