Diagnostic Verification of Atmospheric Water Cycle Predicted by Regional Mesoscale Models and Ensemble Systems

Abstract

Precipitation is the final component of a complex process chain of the atmospheric water cycle. All model errors in this process chain are consequently accumulated in quantitative precipitation forecasts (QPF). To diagnose the shortcomings of QPF, the following four key variables of the atmospheric water cycle have been evaluated: integrated water vapour content (IWV), low cloud cover (LCC), high cloud cover (HCC), and precipitation rate at the surface. This comprehensive verification of all key variables is performed for nine deterministic models and four ensemble systems from the forecast demonstration experiment of Mesoscale Alpine Program (MAP D-PHASE) using measurements from the General Observation Period (GOP) over Southern Germany for summer 2007. Verification of individual key variables reveals that most of the models forecast the mean values of IWV very well; however, they show large biases in the mean values of LCC, HCC, and precipitation. At certain times and locations, all models show large errors in all key variables, especially in HCC and precipitation. The models with convection parameterization predict diurnal precipitation maxima a few hours earlier than observations, whereas deep-convection-resolving models forecast the diurnal maxima too late. Early initiation of convection is a specific problem of the Tiedtke convection scheme. The forecast performance of high resolution models is superior to their corresponding low resolution models for all key variables, except for IWV. Multivariate verification fails to quantify the shortcomings in QPF, perhaps due to the limited availability of observations. Multimodel multiboundary ensemble prediction systems (EPS) show superiority in the prediction of all key variables and also has better representation of forecast uncertainty compared to EPS based on a single model. EPS which accounts the small-scale perturbations, due to the uncertainty in boundary and initial conditions from limited area models, lead to better forecasts for strong events. However, all the EPS evaluated in this study are underdispersive which clearly implies that they are not able to account for all possible uncertainties of short-range forecasts.