LAUSR

Abstract The study presents a new method nowcasting precipitation called the Ensemble Tree Method (ETM), which gives probability forecast of accumulated precipitation based on the extrapolation of radar reflectivity. ETM combines a tree model with a Bootstrap technique. It forecasts the probability that the hourly accumulated precipitation exceeds a given threshold for cells of 3 by 3 km size. ETM was tested using radar reflectivity data from July 2012 in a domain of 489 km by 291 km covering the Czech Republic (Central Europe). While forecasting, we considered a lead time of up to 180 min having a time step of 30 min and four precipitation thresholds (0.1, 1.0, 5.0, and 10.0 mm). ETM provided us forecasts of the probability of exceeding an hourly precipitation threshold from 0 to 60 min, 30 to 90 min, …, and 120 to 180 min. The performance of ETM was assessed using a skill score derived from the mean-square-error, and was compared with the performance of forecasts based on a logistic regression that was used as reference forecast. We demonstrated that the prediction of ETM is better than that of the reference forecast. The main advantage of ETM is that the ETM reflects the uncertainty of forecast better as compared to the overconfident reference forecasts, which is particularly true for the higher precipitation thresholds. Thus, despite low predicted probabilities, the forecasts given by ETM seem more realistic.