A new verification method to ensure consistent ensemble forecasts through calibrated precipitation downscaling models

A new verification method is proposed to test the consistency of ensemble high-resolution precipitation fields forecasted by calibrated downscaling models. The method is based on a generalization of the verification rank histogram and tests the exceedance probability of a fixed precipitation threshold calculated from the observed or ensemble fields. A graphical tool that accounts for random assignments of the rank is proposed to provide guidance in histogram interpretation and to avoid a possible misunderstanding of model deficiencies. The verification method is applied on three numerical experiments carried out in controlled conditions using the space-time rainfall (STRAIN) downscaling model with the aims of investigating (i) the effect of sampling variability on parameter estimation from the observed fields and (ii) model performance when calibration relations between the parameter and a coarse meteorological observable are used to interpret events arising from one or more physical conditions. Results show that (i) ensemble members generated using the parameters estimated on the observed event are overdispersed; (ii) the adoption of a single calibration relation can lead to the generation of consistent ensemble members; and (iii) when a single calibration relation is not able to explain observed event variability, storm-specific calibration relations should be adopted to return consistent forecasts.