A new framework for probabilistic seasonal forecasts based on circulation type classifications and driven by an ensemble global model

Abstract

Abstract. In the last years coupled atmospheric ocean climate models have remarkably improved medium range seasonal forecasts, especially on middle latitude areas such as Europe and the Mediterranean basin. In this study a new framework for medium range seasonal forecasts is proposed. It is based on circulation types extracted from long range global ensemble models and it aims at two goals: (i) an easier use of the information contained in the complex system of atmospheric circulations, through their reduction to a limited number of circulation types and (ii) the computation of high spatial resolution probabilistic forecasts for temperature and precipitation. The proposed framework could be also useful to lead predictions of weather-derived parameters, such as the risk of heavy rainfall, drought or heat waves, with important impacts on agriculture, water management and severe weather risk assessment. Operatively, starting from the ensemble predictions of mean sea level pressure and geopotential height at 500 hPa of the NCEP – CFSv2 long range forecasts, the third-quantiles probabilistic maps of 2 m temperature and precipitation are computed through a Bayesian approach by using E-OBS 0.25∘ gridded datasets. Two different classification schemes with nine classes were used: (i) Principal Component Transversal (PCT9), computed on mean sea level pressure and (ii) Simulated Annealing Clustering (SAN9), computed on geopotential height at 500 hPa. Both were chosen for their best fit concerning the ground-level precipitation and temperature stratification for the Italian peninsula. Following this approach an operative chain based on a very flexible and exportable method was implemented, applicable wherever spatially and temporally consistent datasets of weather observations are available. In this paper the model operative chain, some output examples and a first attempt of qualitative verification are shown. In particular three case studies (June 2003, February 2012 and July 2014) were examined, assuming that the ensemble seasonal model correctly predicts the circulation type occurrences. At least on this base, the framework here proposed has shown promising performance.