A comparison between subjective and objective thunderstorm forecasts

We made a comparison between man-made 24-h thunderstorm forecasts (subjecti ve forecasts) and 24-h pure index thunderstorm forecasts (objective forecas ts). The forecasts concern the plane of Friuli-Venezia Giulia region (NE It aly) during summer. Subjective forecasts are based on forecaster interpreta tion of numerical models outputs and forecaster's experience, while objecti ve forecasts use models outputs and the thunderstorm climatology of our reg ion to issue two independent indexes. Both these indexes are calculated by a linear combination of meteorological variables extracted from the Europea n Centre for Medium Range Forecast (ECMWF) model. In one case, the coeffici ents of the combination were obtained by a linear multiregression over 8 ye ars of data, while in the other, the biserial correlation coefficient was m aximized on a subset of 2 years. To estimate the quality of forecasts, we u sed the data collected during the summer of 1998 and 1999; thunderstorm day s are defined by lightning data and ground stations observations. The joint probability density functions of forecasts and observations have been used to study the forecast quality. We also studied the reliability diagrams of subjective forecasts. We found that subjective forecast skill and objectiv e forecast skill are comparable although there were some differences as wel l. Subjective forecasts have a low false alarm rate (FAR) and they showed g ood facility in forecasting days without thunderstorms, but they score low in probability of detection (POD). Objective forecasts, in contrast, showed lower FAR and facility in forecasting stormy days but detected thunderstor ms better than subjective forecasts. There are significant fluctuations in forecasts' performance from 1998 to 1999 summer seasons. This is likely due to variation in thunderstorm climatology and thunderstorm type frequencies . Furthermore, feedback of past performances could have played an important role in subjective forecasts. (C) 2001 Elsevier Science B.V. All rights re served.