Probabilistic Forecasts of Precipitation in Terms of Quantiles Using NWP Model Output

Abstract At sites with observations it is often possible to improve or enrich NWP model forecasts by means of statistical methods. Such forecasts are almost exclusively deterministic or probabilities of discrete events. In this paper a flexible approach for making reliable precipitation forecasts in terms of quantiles is described. The approach is essentially in two steps: (i) estimation of probability of precipitation and (ii) estimation of selected quantiles in the distribution of precipitation amounts given occurrence of precipitation. Estimates are obtained by means of probit regression and local quantile regression, respectively. By applying the laws of probability, the steps are combined to make unconditional quantile forecasts. Examples of daily precipitation forecasting using single deterministic forecasts and ensemble forecasts as input are given.