PubAg

Main content area

Improving ECMWF-based 6-hours maximum rain using instability indices and neural networks

Author:
Manzato, Agostino, Pucillo, Arturo, Cicogna, Andrea
Source:
Atmospheric research 2019 v.217 pp. 184-197
ISSN:
0169-8095
Subject:
lightning, neural networks, rain, Italy
Abstract:
Friuli Venezia Giulia (FVG, NE Italy) is an area of maximum rainfall in the whole Alpine chain territory, reaching more than 3200 mm of mean annual rain in the Julian Prealps. According to recent climatological studies, the same area is also one of the European spot in recent lightning climatologies, meaning that convective rain plays an important role in the total rainfall.A network of 104 raingauges placed around the FVG territory is used to extract the absolute maximum rain accumulated every 6 hours in four subareas of FVG. In an attempt to improve the original ECMWF maximum rain, these data have been targeted to develop 32 statistical downscaling models, according to the period of the day, of the year and specific sub-area. ECMWF 6-hour rain forecasts available for all the gridpoints encompassed in the FVG territory and some derived variables (absolute values, anomalies, standardized values, plus mean, max and SD in time and/or space) have been used as predictors.With respect to a previous version of this work, here also the instability pseudo-indices (derived from the vertical profile with the maximum vertical resolution available in the ECMWF hybrid levels) are used as candidate predictors. Moreover, also non-linear methods, namely neural networks, are implemented, together with exhaustive multiregression models. Results show that the 32 models improve -on average- R2 of 12% on the validation sample and of 5% on the 2017 test sample, with respect to the ECMWF rain forecast, but the improvement is particularly notable during the convective season (18%).
Agid:
6228056