Verification of Precipitation Forecasts from Two Limited-Area Models over Italy and Comparison with ECMWF Forecasts Using a Resampling Technique

Abstract

Abstract This paper presents the first systematic limited area model (LAM) precipitation verification work over Italy. A resampling technique was used to provide skill score results along with confidence intervals. Two years of data were used, starting in October 2000. Two operational LAMs have been considered, the Limited Area Model Bologna (LAMBO) operating at the Agenzia Regionale Prevenzione e Ambiente-Servizio Meteorologico Regionale (ARPA-SMR) of the Emilia–Romagna region, and the QUADRICS Bologna Limited Area Model (QBOLAM) running at the Agenzia per la Protezione dell’Ambiente e per i Servizi Tecnici (APAT). A 24-h forecast skill score comparison was first performed on the native 0.1° high-resolution grids, using a Barnes scheme to produce the observed 24-h accumulated rainfall analysis. Two nonparametric skill scores were used: the equitable threat score (ETS) and the Hanssen and Kuipers score (HK). Frequency biases (BIA) were also calculated. LAM forecasts were also remapped on a lower-resolution grid (0.5°), using a nearest-neighbor average method; this remapping allowed for comparison with ECMWF model forecasts, and for LAM intercomparisons at lower resolution, with the advantage of reducing the skill score sensitivity to small displacements errors. LAM skill scores depend on the resolution of the verification grid, with an increase when they are verified on a lower-resolution grid. The selected LAMs have a higher BIA compared to ECMWF, showing a tendency to overforecast precipitation, especially along mountain ranges, possibly due to undesired effects from the large-scale and/or convective precipitation parameterizations. Lower ECMWF BIA accounts for skill score differences. LAMBO precipitation forecasts during winter (adjusted for BIA differences) have less misses than ECMWF over the islands of Sardinia and Sicily. Higher-resolution orography definitely adds value to LAM forecasts.