Articles | Volume 12, issue 3
https://doi.org/10.5194/nhess-12-591-2012
https://doi.org/10.5194/nhess-12-591-2012
Research article
 | 
08 Mar 2012
Research article |  | 08 Mar 2012

Intercomparison of two meteorological limited area models for quantitative precipitation forecast verification

E. Oberto, M. Milelli, F. Pasi, and B. Gozzini

Abstract. The demand for verification of numerical models is still very high, especially for what concerns the operational Quantitative Precipitation Forecast (QPF) used, among others, for evaluating the issuing of warnings to the population. In this study, a comparative verification of the QPF, predicted by two operational Limited Area Models (LAMs) for the Italian territory is presented: COSMO-I7 (developed in the framework of the COSMO Consortium) and WRF-NMM (developed at NOAA-NCEP). The observational dataset is the precipitation recorded by the high-resolution non-GTS rain gauges network of the National Civil Protection Department (NCPD) over two years (2007–2008). Observed and forecasted precipitation have been treated as areal quantity (areal average of the values accumulated in 6 and 24 h periods) over the 102 "warning areas", defined by the NCPD both for administrative and hydrological purposes. Statistics are presented through a series of conventional indices (BIAS, POD and POFD) and, in addition, the Extreme Dependency Score (EDS) and the Base Rate (BS or 1-BS) have been used for keeping into account the vanishing of the indices as the events become rare. Results for long-period verification (the whole 2 yr) with increasing thresholds, seasonal trend (3 months period), diurnal error cycle and error maps, are presented. Results indicate that WRF has a general tendency of QPF overestimation for low thresholds and underestimation for higher ones, while COSMO-I7 tends to overestimate for all thresholds. Both models show a seasonal trend, with a bigger overestimation during summer and spring, while during autumn and winter the models tend to be more accurate.

Download
Altmetrics