1. A study on the spread/error relationship of the COSMO-LEPS ensemble Purpose of the work  The spread-error spatial relationship is good, especially after the 2nd forecast-day (the spread plays a predictive role in the geographical distribution of forecast skill)  The spread-error relationship varies with the geographical position  The spread-error relationship in time (not shown) needs a larger population in order to be measured  Differences day/night: both spread and error decrease at night (not shown) Conclusions Michele Salmi 1, Chiara Marsigli 2, Andrea Montani 2, Tiziana Paccagnella 2 1 Department of Physics, University of Ferrara, Italy 2 ARPA-SIMC, HydroMeteoClimate Service of the Emilia-Romagna, Bologna, Italy Results Data and method COSMO-LEPS METHODOLOGY COSMO-LEPS (Limited-area Ensemble Prediction System of the COSMO Consortium) is an operational mesoscale ensemble, running daily since November 2002, at 12 UTC. The domain covers large part of Central and Southern Europe, with a spatial resolution of 7 km (from Dec 2009). COSMO-LEPS has been designed to provide probabilistic predictions of surface parameters, especially for severe weather, in late-short to early-medium range (up to day 5.5). COSMO-LEPS is a downscaling of the ECMWF EPS, where 16 EPS members are selected and used to provide initial and boundary conditions to the COSMO runs. A set of COSMO model physics parameters are also perturbed.  COSMO-LEPS operational forecasts:  horizontal resolution 10 km  forecasts are interpolated on a 0.25x0.25 deg regular lat-lon grid (the same of the ECMWF data)  ECMWF analysis:  ECMWF operational analyses (horizontal resolution of approx. 25 km) are extracted over a regular lat-lon grid (0.25x0.25 deg)  Ensemble Spread: the COSMO-LEPS spread is defined as the ensemble standard deviation, the RMS distance of each member from the ensemble mean  Ensemble Error: the COSMO-LEPS forecast error is the RMS error of the ensemble mean with respect to the ECMWF analysis  The evaluation is performed for summer and fall 2009. This work aims at providing an evaluation of the statistical properties of the COSMO-LEPS ensemble. In particular, the relationship between the spread and the forecast error of the COSMO-LEPS ensemble is here evaluated, with the purpose of quantifying the capability of the ensemble of representing the forecast error, in a statistical sense. The analysis is performed for two upper-air meteorological variables: geopotential height at 700 hPa and temperature at 850 hPa. The dependence of this relation from the forecast range, from the season, and from the spatial location is also assessed. Dependence on season and forecast range Before computing the spread/error relationship, a orography mask is turned on (>1000m), in order to reduce the spurious error due to mountains. Geographical dependence The curved form of the spread/skill lines indicates that in the central spread categories the error and the spread are uncorrelated, i.e. a wide range of error values correspond to specific spread values. BOX SW BOX NWBOX NE BOX SE Hence, the spatial domain has been split in 4 areas of equal size, in order to highlight possible dependencies of the relation from the location. A dedicated analysis of the spread/skill relationship have been performed over Italy, though excluding the Alpine area due to the detrimental effect of the mountains on the T850 field. The SE area has the best spread/error correspondence, followed by the NW one, especially for the summer season. Instead, the NE area is largely responsible of the apparent lack of correlation for the intermediate spread values, together with the SW area. In addition, in summer the relations are generally better, while in autumn the problem is more evident. This results indicate a strong dependence of the ability of the ensemble to forecast the forecast error on geographical location and season and, hence, on the typical meteorological situations. AREA IT Over the selected domain, the relation between spread and error is more linear than over the full domain. Contact e-mail: cmarsigli@arpa.emr.it