Heavy Rainfall Identification within the Framework of the LEXIS Project: The Italian Case Study

LEXIS (Large-scale EXecution for Industry and Society) H2020 project is currently developing an advanced system for Big Data analysis that takes advantage of interacting large-scale geographically-distributed HPC infrastructure and cloud services. More specifically, LEXIS Weather and Climate Large-Scale Pilot workflows ingest data coming from different sources, like global/regional weather models, conventional and unconventional meteorological observations, application models and socio-economic impact models, in order to provide enhanced meteorological information at the European scale. In the framework of LEXIS Weather and Climate Large-scale Pilot, CIMA Research Foundation is running a 7.5 km resolution WRF (Weather Research and Forecasting) model with European coverage, radar assimilation over the Italian area, and daily updates with 48 hours forecast. WRF data is then processed by ITHACA ERDS (Extreme Rainfall Detection System - http://erds.ithacaweb.org), an early warning system for the monitoring and forecasting of heavy rainfall events. The WRF model provides more detailed information compared to GFS (Global Forecast Systems) data, the most widely used source of rainfall forecasts, implemented in ERDS also. The entire WRF - ERDS workflow was applied to two of the most severe heavy rainfall events that affected Italy in 2020. The first case study is related to an intense rainfall event that affected Toscana during the afternoon and the evening of 4th June 2020. In this case, the Italian Civil Protection issued an orange alert for thunderstorms, on a scale from yellow (low) to orange (medium) to red (high). In several locations of the northern part of the Region more than 100 mm of rainfall were recorded in 3 hours, corresponding to an estimated return period equal to or greater than 200 years. As far as the 24-hours time interval concerns, instead, the estimated return period decreases to 10-50 years. Despite the slight underestimation, WRF model was able to properly forecast the spatial distribution of the rainfall pattern. In addition, thanks to WRF data, precise information about the locations that would be affected by the event were available in the early morning, several hours before the event affected these areas. The second case study is instead related to the heavy rainfall event that affected Palermo (Southern Italy) during the afternoon of 15th July 2020. According to SIAS (Servizio Informativo Agrometeorologico Siciliano) more than 130 mm of rain fell in about 2.5 hours, producing widespread damages due to urban flooding phenomena. The event was not properly forecasted by meteorological models operational at the time of the event, and the Italian Civil Protection did not issue an alert on that area (including Palermo). During that day, in fact, only a yellow alert for thunderstorms was issued on northern-central and western Sicily. Within LEXIS, no alert was issued using GFS data due to the severe underestimation of the amount of forecasted rainfall. Conversely, a WRF modelling experiment (three nested domain with 22.5, 7.5 and 2.5 km grid spacing, innermost over Italy) was executed, by assimilating the National radar reflectivity mosaic and in situ weather stations from the Italian Civil Protection Department, and it resulted in the prediction of a peak rainfall depth of about 35 mm in 1 hour and 55 mm in 3 hours, roughly 30 km far apart the actual affected area, thus values supportive at least a yellow alert over the Palermo area. Obtained results highlight how improved rainfall forecast, made available thanks to the use of HPC resources, significantly increases the capabilities of an operational early warning system in the extreme rainfall detection. Global-scale low-resolution rainfall forecasts like GFS one are in fact widely known as good sources of information for the identification of large-scale precipitation patterns but lack precision for local-scale applications.