Location errors in QPFs over the Calabria region: Does a multi-model poor man's ensemble over-perform each member?

In this paper we compare the location errors of precipitation patterns forecast by three models in two years of operational activity with those obtained by combining together the three single forecasts to obtain poor's man ensembles. Aim of the work is check whether precipitation forecasts obtained by ensembles can be better, in term of location errors, of single models on the Calabria region. Four were the strategies adopted for combining precipitation forecasts, namely the average, the median, the probabilistic matching using with average and the median. The Contiguous Rain Area (CRA) analysis was the method to find out the location errors of forecast precipitation patterns. The ability of models to correctly forecast the location errors was measured by means of an index, called the CRA mean shift (CMS) index that summarizes the CRA outcomes, which are a lot of couples of numbers each of them representing longitude and latitude for each event considered. The comparison was performed on statistical base by bootstrapping the coordinated of the location errors of the single models and the ensembles. Spite of a general improvement of ensemble forecasts that resulted in a lower CMS index with respect to single models, such improvement was not significant for all ensembles. In particular when models showed large location errors, obtained when the CRA best-fit criterion the minimization of mean square error, the best result of one out of four combination strategies, where the ensemble median was chosen as forecast, was significantly (at 95%) better than two single models.