MOS-Based Precipitation Forecasts for River Basins

Abstract

Statistical interpretation models of the Aire Limitée Adaptation Dynamique Développement International/Limited Area Modelling in Central Europe (ALADIN/LACE) numerical weather prediction (NWP) model outputs have been developed to improve both quantitative and probabilistic precipitation forecasts (QPF and PQPF, respectively) for the warm season. Daily means (from 0600 to 0600 UTC of the next day) of area precipitation are forecast for seven river basins by using the prognostic fields of the NWP model, which began the integration at 0000 UTC. The selected river basins differ in size and mean elevation above mean sea level. A dense network of rain gauges, where the mean distance between the two nearest neighbors is about 8 km, is used to calculate basin average precipitation amounts. Data from three warm seasons (April?September 1998?2000) were used to develop and verify statistical interpretation models. Several statistical models based on multiple linear regression were used to produce QPF and were compared. They estimated either the direct value of the areal mean precipitation or the difference between the NWP model forecast and the actual value. The statistical models also differed in the training data used to develop model parameters. Two different statistical models, multiple linear regression and logistic regression, were used to produce PQPF, and their performances were compared. Model output statistics (MOS) was used to find suitable predictors and to calculate model coefficients. MOS was applied to two seasons; the remaining one served as the independent verification dataset. All three combinations of seasons were considered. The statistical interpretation models significantly improved both the QPF and PQPF of the NWP model forecast. The root-mean-square error of the QPF from the direct NWP model forecast decreased by about 10%?30% for individual river basins.