6-hour maximum rain in Friuli Venezia Giulia: Climatology and ECMWF-based forecasts

• 104 raingauges are used to extract the absolute maximum rain accumulated every 6h in FVG (NE Italy) during 9 years.
• The climatology of 6-haur max rain is studied and then the comparison with ECMWF forecasts are performed.
• To improve these forecasts 32 linear regression models have been developed.

Friuli Venezia Giulia (FVG) is a region in Italy with very complex orography, having an annual rainfall amount that varies from about 900 mm on the coast to more than 3200 mm in the Julian Prealps. A network of 104 raingauges placed around the FVG territory was used to extract the absolute maximum rain accumulated every 6 h, during the period 16 February 2006 to 15 February 2015 (9 years). Interannual, annual, weekly and daily cycles of three classes of rain intensities are analyzed, finding that significant rainfalls (MaxRain > 5 mm) are more frequent in the May to mid-August period, while the heaviest rainfalls (> 40 mm) are more probable between May and the beginning of December, with a peak at the very beginning of November.ECMWF 6-h forecasts at 18 gridpoints (spaced at 0.25°) above the FVG region are studied for the same period, to find the maximum 6-h rain forecasted by the ECMWF model from + 6 to + 48 h and correlate it with the observed maximum rain of all the 104 raingauges. It is found that the correlation coefficient R is higher at 0000–0600 UTC and minimum at 1800–0000 UTC, while the BIAS is always negative (underestimation), varying between − 3.5 and − 6.9 mm. Looking at more homogeneous subareas, ECMWF has a much worse BIAS and RMSE for the Prealps zone, while its correlation coefficient is lower for the coastal and plains zones.For comparison, a similar exercise is repeated using a LAM model (ALADIN-ARSO), finding better BIAS and RMSE, but a lower skill for the mean correlation coefficient. Hence, a linear statistical method (multiregression with exhaustive input selection) for forecasting the maximum 6-h rain using as candidate predictors the direct model output (absolute values, anomalies, standardized values, plus mean, max and SD in time and space) is developed independently for four different sub-regions and two periods of the year starting from the ECMWF forecast. It is found that the strong BIAS in the Prealpine area can easily be removed, substantially improving the forecast, in particular during the October–April period, while the plains and coastal area, in particular during May–September, have the lowest predictability.