A multi-model/multi-analysis limited area ensemble: calibration issues

In this paper four different post-processing techniques: the Bayesian model averaging (BMA), the ensemble model output statistics (EMOS) with a variant known as EMOS+ and a new dressing kernel (DRESS) are applied and compared, in a pre-operational context, to calibrate a mesoscale multi-model multi-analysis ensemble. The ensemble makes use of three different limited area models (Bologna Limited Area Model (BOLAM), MM5 and RAMS), one of them used twice with different setups, fed with two sets of analysis and boundary conditions obtained from the European Centre for Medium-Range Weather Forecasts (ECMWF) and National Centers for Environmental Prediction (NCEP) general circulation models. The resulting ensemble of eight members was run for a period of 6 months (from October 2002 to April 2003) in a Euro - Atlantic domain. The forecast was validated against 2 m temperature measured at 21 meteorological stations scattered across Sardinia (Italy). For each method the calibration ability was assessed evaluating the flatness of the rank histogram, the coverage of the expected forecast intervals and the width of the associated probability distribution function. Results show that BMA and DRESS are the best in improving the calibration of the raw ensemble whereas EMOS and EMOS+ have proven worse, with the latter marginally better. Copyright (c) 2008 Royal Meteorological Society.