Factors affecting the quality of QPF: a multi‐method verification of multi‐configuration BOLAM reforecasts against MAP D‐PHASE observations

This paper discusses the results of a quantitative precipitation forecast ( QPF ) verification study aimed at comparing different alternative new configurations of the meteorological BOlogna L imited A rea M odel ( BOLAM ) to be implemented in 2013 into the ISPRA 's H ydro‐ M eteo‐ M arine forecasting system ( SIMM ) in place of the currently operational version, dated 2009. Five new configurations are defined after considering several combinations of the following model settings: horizontal grid spacing, domain extension, initial and boundary conditions, nesting design and the BOLAM code version. Such testing configurations are inter‐compared with the operational BOLAM version following a multi‐method approach including F ourier spectral analysis, several categorical scores, quasi‐relative operating characteristic diagram and visual inspection of the spatial distributions of the contingency table elements. Rain gauge measurements available between J une and N ovember 2007 within the international initiative ‘ D emonstration of P robabilistic H ydrological and A tmospheric S imulation of flood Events in the A lpine region’ of the M esoscale A lpine P rogramme ( MAP D‐PHASE ) are considered as observational dataset. A 6 month reforecast campaign is performed in order to produce, for each compared configuration, the corresponding forecast series. Results show that decreasing model grid spacing and simultaneously increasing model domain extension is effective in improving the QPF quality when higher‐resolution initial and boundary conditions are directly applied to BOLAM , without using a coarser‐resolution parent model run. Therefore, such skilful configuration has been deployed in late 2012 within the international research programme ‘ HYdrological cycle in M editerranean EXperiment ’ ( HyMeX ). On the contrary, when keeping the low‐resolution, double nesting configuration, improving input data decreases the QPF performance.