Stochastic temporal downscaling in Northeast Italy using convection-permitting climate models: from hourly to sub-hourly timescales

The statistical properties of rainfall at short durations are pivotal for many hydrological applications. Commonly available rainfall records nor km-scale model, i.e. Convection-Permitting Models (CPMs), do not provide rainfall data at the sub-hourly scales needed for many applications, such as hydrological modelling in small or urban catchments or landslide or debris-flow models. Motivated by the above considerations, in this application a statistical downscaling technique is proposed for inferring the rainfall correlation structure at sub-hourly scale by using hourly statistics from CPM simulations. The proposed approach is based on the theory of stochastic processes, which establishes statistical relationships between coarse-scale predictors and fine-scale predictands. To validate the temporally downscaled results against observations, here we use, as a benchmark, high-resolution rainfall records from a dense network of rain gauges in northeastern Italy considering aggregation timescales ranging from 5 minutes to 24 hours. We then explore how the downscaling method developed here, coupled with the Complete Stochastic Modelling Solution (CoSMoS; Papalexiou, 2018) framework, may be used to generate sub-hourly rainfall sequences that reproduce the observed short- and long-timescale variability. Applied to statistics for each month in a year, to reproduce seasonality, the proposed downscaling method appropriately reproduces the observed correlation structure at desired fine-scale resolution. Consequently, the rainfall generator used here, by exploiting the downscaled information from CPM runs, allows to generate rainfall records at the desired scale that may be used for evaluating risk and risk change scenarios, for example associated with debris flows.