We examine a method for predicting variations in the probabilities and occurrence of intense local 24-hour precipitation events over seasonal, annual, and 5-yr intervals. The wet-day mean μ can be used to describe the exponential distribution for the wet-day amounts, and provides a basis for forecasting the likelihood of seeing an event above a given threshold. A regression-based downscaling model is calibrated on annualμ and wet-day frequency fw, respectively, and then tested on seasonal to multi-annual scales. The probabilities for heavy precipitation statistics are taken as the product between a fitted cumulative exponential distribution for the wet-day amounts μ and fw. The annual number of heavy precipitation events is estimated from the annual probability, where a 90% confidence interval on the number is computed using the 5 and 95 percentiles of a binomial distribution for the predicted probability. The analysis identifies a strong link between large-scale predictors such as mean sea-level pressure or surface temperature and the wet-day frequency. There was a weaker dependency of the wet-day mean to large-scale predictors, which suggests that local-scale processes have a stronger influence on the 24-hour precipitation amounts. The results also suggest that similar physical processes influence the wet-day mean μ on different timescales except for during winter, and that similar physical processes may explain the wet-day frequency on seasonal to annual time scales. The implication is that models calibrated on annual data samples in some cases can give skilful predictions on seasonal time scales.
Keywords: Precipitation, wet-day mean, wet-day frequency, seasonal-to-decadal forecast, validation, statistical downscaling