Italy can boast of a role at the highest level in the development of meteorological observations , with 6 meteorological stations operating since the 18th century (Bologna, Milano, Roma, Padova, Palermo and Torino), and 15 stations with observation starting in the first half of the 19th century. The first attempts of performing a systematic collection of monthly rainfall data go back to as early as 1880 when the National Office for Meteorology and Climate was founded. The National Hydrographic Service (SIN) and the National Hydrographic and Mareographic Service (SIMN) collected annual maxima values for 1, 3, 6, 12 and 24 h durations in the Hydrological Yearbooks from 1917 to early 2000s (the final publication year depends on the local agencies of the SIMN). The D.Lgs 112/1998 dismantled the SIMN, transferring its tasks to the 19 administrative regions and the 2 autonomous provinces of Trento and Bolzano. These authorities were designated as local Operational Centres and Regional Environmental Agencies to deal with hydro-meteorological monitoring and civil protection issues.

In spite of the huge heritage of data, only a small fraction of the Italian rainfall data is available in a computer-readable format. Moreover, the dismantlement of the National Service led to a lack of updates for the national database of extreme rainfall that is still stuck, for some regions, at the beginning of the 1990s. This has led to a very fragmented framework: updated rainstorm hazard assessments are actually only available for some regions and only at the regional scale (see, e.g. ). Various regional studies present different methodologies and are sometimes based on very different data densities and record lengths e.g., but few updated analyses on short-duration rainfall in a regional framework are available e.g..

In view of the assembling of the first comprehensive dataset of extreme rainfall of short duration in Italy several major sources of data have been analysed. The resulting dataset, referred to as the Italian Rainfall Extremes Database (I-RED), includes data from more than 4500 stations across the country, spanning the period between 1916 and 2014, and refers to annual maximum rainfall recorded in 1 to 24 consecutive hours (exact durations available are 1, 3, 6, 12 and 24 h).

The following sections describes the sources of the data, the work carried out for the merging of the database and the operations that are still required for making it suitable for nationwide robust rainfall frequency analyses. A preliminary analysis of the extreme rainfall regime at the national scale is also presented.

The number of data available per year in the I-RED is reported in Fig. a, compared with that of the CUBIST database. As every station is related to a unique value of annual maxima for a given duration, the presence of a measurement implies the presence of a station. The number of available stations increases with time, and drastically grows after the dismissal of the SIMN and the development of the local agencies. The decrease after 2010 can be attributed to the fact that not all the regions have published the data for the most recent years.

The smaller size of the I-RED compared the CUBIST database in some years can be due to the following:

The presence, before 1945, in the CUBIST database of data from territories lost by Italy after World War II (e.g. Istria) or from neighbouring countries, not included in the I-RED;

For a descriptive analysis of the rainfall data, all the assembled time series are classified according to their length. Results are shown in Fig. b. Considering the short life of the rain gauges installed by the regional operational centres, a large percentage of the series is shorter than 20 years but the contribution of the CUBIST database allows for a significant amount of longer series. The series with more than 80 years of data for the 1, 3, 6, 12, 24 h durations are 16, 14, 15, 14 and 17 respectively. In general, all the durations report a similar behaviour, despite some differences in the distribution of the null values as shown in Fig. c. The reasons that lead to missing data only for certain durations can be various and related to either the measuring, the recording or the storage of the data (e.g. missed reading of the record from the operator, data classified as not-valid in the validation phase).

The spatial distribution of the stations is shown in Fig. d. The colour scale refers to the number of the available data per series. The minimum number across the five durations is considered. One can clearly distinguish that, even if the whole national territory is represented, the density of the stations widely changes across the nation. To show the relevance of the non-uniformity, a gridded domain with a mesh size of 50 km is introduced. Figure  shows the total number of data per cell, i.e., in the sum across the whole period of the annual number of stations with available data falling in the cell. If data consistency changes for the different durations, the shortest one is considered. The non-uniformity of the network density clearly emerges, with some cells presenting almost 10 times the number of data of other cells. The most densely gauged cells can be found in the north-west of the country, in particular in the Liguria region, in northern Tuscany and in the north-east.

A preliminary descriptive analysis of the characteristics of extreme rainfalls at the national scale has been carried out on the newly developed I-RED database. Series with a minimum length of 20 years of data have been considered in this analysis. This length constraint leads to a subset of 1974 series available for the analysis, out of the original 4686. For each duration, the median of the series is depicted in Fig. . The median is used as a robust estimator of the central tendency of a series, less sensitive than the mean to the presence of outliers. As common methods of fitting distributions (e.g. product moments or L-moments) use mean values for representing the central tendency, maps of the mean for the different durations are attached in the Supplement.

Some geographical areas are characterized by clusters of large median values and these clusters appear consistent across the different durations. Furthermore, at the country-wide scale we observe that the coefficient of variation of the medians increases for increasing durations, suggesting a wider range of variability of the corresponding median values.

For each series, the sample L-moments have then been computed to describe the shape of the empirical distribution of the records. The mean L-moment ratios among the different durations give information respectively on the dispersion (L-CV), skewness (L-skewness) and “peakedness” (L-kurtosis) of the empirical distributions. All the above statistics are mapped in Fig. . Considering that the L-moment ratios show similar behaviour for the considered durations, we decided for simplicity to include only the average ones in the paper. The maps for the different durations are reported in the Supplement. Figure f shows that the coastal areas and the islands are generally characterized by a higher variability in the annual maxima series, presenting larger L-CV values. The northern part of the peninsula, even if characterized by large median values, shows lower L-CV, which is typical of areas with large average rainfall values. It is harder to identify a precise spatial pattern in the distribution of the skewness and kurtosis values (Fig. g and h). Coastal and island areas seem to generally show larger skewness values, confirming the influence of the Mediterranean Sea on the climate of these areas. All the aforementioned maps have been also interpolated for visualization purposes with ordinary kriging; detailed results are reported in the Supplement.

The significance of the developed dataset also allows preliminary exploration of the rainfall events sometimes referred to as “black swans” , showing extraordinary intensities even when compared with the population of annual maxima. In Italy, many of these events have been studied as individual extraordinary events e.g., due to the large recorded intensities and to their severe consequences, but the fragmented configurations of the national database have prevented a systematic treatment of this population of “extremes of the extremes” . A preliminary investigation on the occurrence of very extreme events at the national scale has been performed and summarized in Fig. a, which shows the spatial distribution of the record-breaking rainfall events for the considered durations from 1935 to 2015. A record-breaking event is defined as the annual value that exceeds all the previous ones. At this stage, only nationwide records are considered, pulling up all the data together year by year. Record-breaking rainfall amounts can provide a picture of the spatio-temporal distribution of the major weather anomalies in the country. Analysing record-breaking events has some advantages from both an operational and a statistical point of view. Due to the significant amounts recorded, these events can be easily verified by combining different sources of information, and, moreover, this kind of analysis does not require any assumption on the underlying probability distribution . The spatial distribution of the events seems to suggest a clustering of these phenomena in some areas of the country: the eastern part of Liguria and northern part of Tuscany and the extreme south of Calabria. Localized events also occurred in Campania, Sicily and Sardinia. All of these areas are generally characterized by complex orography in the proximity of the coastline: a framework that can promote the development of particularly intense phenomena . On-site systematic analysis of the record-breaking events, which are expected to provide useful information for characterizing the extreme rainfall regime in the country , is now possible thanks to the consistency of the new I-RED database. Figure b shows the record-breaking evolution over time (for each duration); the occurrence of a new records is represented by an increasing step in the line. Figure c reports the same records, whose values are normalized by the 1935 values.

The first comprehensive dataset of extreme rainfall in Italy, called I-RED, has been presented here. It is a significant source of information, able to provide unprecedented knowledge on the characteristics of heavy precipitation in Italy and on the possible rainfall regime changes in the last century. Further efforts will be addressed to increase the spatial data homogeneity and coverage in time, by including the data of the most recent years and, eventually, by contacting the local authorities for requesting assistance in the merging of the series. The final aim is to make the update of the database systematic and unsupervised. This can be done by strengthening the collaboration with the data providers, in the framework of joined projects, as did the one that led to the development of the ArCIS dataset, collecting updated rainfall and temperature data from a group of regional authorities in northern Italy. Collaborations with other projects, focused on different spatial or temporal scales, will be also explored in order to automatically and efficiently analyse the consistency of the I-RED dataset and to integrate it with the existing ones. A possible target is the SCIA dataset referring to the 24 h and daily scale. Joined projects with international institutions will be evaluated and endorsed in order to make available the I-RED database in larger frameworks for trans-boundary exchange of precipitation data. In the meantime the I-RED will be used for exploring the different outcomes provided by this preliminary analysis, e.g. assessing the influence of the spatial distribution of the stations on the observation of record-breaking extreme events, evaluating the presence of trends in the temporal distribution of the “black swans” and analysing the statistical predictability of these kind of events on such a wide and complex domain.