1. Introduction
Instability indices are often taken as proxies of the probability of storm occurrence and/or intensification (Rasmussen and Blanchard 1998; Brooks et al. 2003; Manzato 2003, 2005, 2012; Groenemeijer and van Delden 2007; Mohr and Kunz 2013; Taszarek et al. 2019, 2020). For example, Taszarek et al. (2017) found that lightning occurrence in Europe is mainly a function of convective available potential energy (CAPE) and that large hail is more likely with high values of low-level moisture, steep lapse rates, and high lifting condensation levels. Kunz et al. (2020) found that shear and storm-relative helicity are very useful in discriminating among hail diameters.
If, in a specific region, a strong climatological relationship is found between environmental parameters and storm-related observations, such as lightning flashes, heavy convective rain, and hail, then it can be assumed that the same relation will hold also in the rest of the world, invoking the “invariance of physical processes.” For example, Brooks et al. (2003) found relationships between proximity-sounding-derived indices and the likelihood of severe weather in the United States and then applied them worldwide. In practice, for the first time, they applied the relationship found for the environmental indices, derived from U.S. proximity soundings, to the indices derived from the “pseudosoundings” of the global NCEP reanalysis. In particular, they found that southern Europe has the greatest likelihood of environments favoring significant severe thunderstorms, particularly over Spain and the ex-Yugoslavian countries. Together with spatial invariance, also temporal invariance of these relationships is often assumed. If a temporal trend of an environmental parameter is found, due to the well-known process of global warming, it should automatically reflect a corresponding change in the storm-related observations. For example, Mohr et al. (2015) developed a logistic model for the potential of hailstorm development in Germany based on environmental parameters derived from a downscaled version of ERA-Interim reanalysis and from insurance hail claims during 1992–2008. Subsequently, they applied this statistical model to the years 2021–50, using the environmental information derived from an ensemble of seven debiased regional climate models.
Simon et al. (2023) have recently studied the trend of cloud-to-ground (CG) lightning encompassing Austria, finding a strong increasing trend over the Alpine region, between 1980 and 2019 (40 years). However, that result was found by fitting the “European Cooperation for Lightning Detection” (EUCLID) observations of lightning data, during only 10 years (2010–19), with a regression model based on ERA5 (Hersbach et al. 2020) environmental parameters. With this statistical model, they created a new dataset of estimated (hindcast) lightning data from 1980 to 2019, from the ERA5 environmental parameters. Their studied domain is included in the larger greater Alpine region (GAR), studied by Manzato et al. (2022a), using the same EUCLID CG observed flash data, for a longer period (2005–19; i.e., 15 years). However, in the overlapping period, the results are not similar. For example, in Manzato et al. (2022a), no statistically significant trend was found for the mean-domain value of observed lightning density during 2005–19, in contrast with the results given by the synthetic time series of Simon et al. (2023) in a subdomain.
Moreover, Taszarek et al. (2021) have found discrepancies between ERA5-derived parameters and corresponding indices derived from real soundings. For example, in midlatitudes, they found that ERA5 simulates a decreasing trend of CAPE and an increase of 0–6-km bulk wind shear, while the radiosoundings show exactly the opposite, i.e., an increasing trend of potential instability due to global warming, as was found also by other works (e.g., Diffenbaugh et al. 2013 for the United States), and a decrease in bulk shear. But can an increase in CAPE really assure an increase in storm frequency and/or storm intensity? Are the assumptions of spatial and temporal invariance of these statistical relationships valid? For example, Ban et al. (2015) found that it is inconsistent to extrapolate the future precipitation scaling from the present-day trend.
This work tries to answer such kinds of questions. In particular, we will test if the relation between environmental indices and storm-related observations is invariant in northeastern (NE) Italy, which is a hotspot for severe weather events (Miglietta et al. 2016; Taszarek et al. 2020; Manzato et al. 2022a). To this purpose, we will assess the trends of sounding-derived thermodynamic and kinetic parameters in the last 31 years. Then, we will consider if the thunderstorm-related observations follow the same trends of the environmental parameters. In fact, it is possible that some of the storm features change in a less straightforward way than could be suggested by simple considerations about the trends of the environmental parameters. Already, Allen (2018) was wondering if the probability of storm initiation (also called “thunderstorm trigger”), or even the physical structure of the storm itself, will be the same in a warmer world, with higher potential instability. The crucial concept of triggering convective initiation (CI) is beyond the scope of the present work, because there is no clear environmental index that encompasses the description of all the possible triggering mechanisms [apart that high values of convective inhibition (CIN) make CI triggers less likely]. Trapp and Hoogewind (2016) found higher levels of CAPE in simulations of a warmer climate, but they did not find a proportional increase in simulated convective events with respect to the control climate. A possible explanation is that there will likely be an increase in CIN (i.e., more negative buoyancy) that should prevent the CI processes. In fact, under global warming, Chen et al. (2020) found that their global simulations foresee an increase in CAPE but also an enhanced CIN over land. Another example is Brooks (2013), where the relationships between large-scale environmental conditions, like CAPE and bulk shear, are related to different kinds of local-scale severe weather events. It was found that climatological simulations indicate an increase in CAPE and a decrease in bulk wind shear (between 0 and 6 km). This results in a more favorable environment for storms, but a less favorable environment for tornadogenesis. This could also possibly lead to a change in the relative frequency of the most intense tornadic and hail events, but signals are highly uncertain.
More generally, Beucler et al. (2021) discovered that statistical meteorological relationships, as learned by neural networks, are not “climate change invariant” and that specific statistical models should be developed trying to solve this issue. For example, they suggest to learn more local and robust relationships between storm-scale convection and the associated synoptic environment. In practice, there is no guarantee that the many processes acting at different temporal and spatial scales that lead to storm development can be “simplified” by the simple statistical relations suggested by correlations among average values.
Manzato (2012) already found that it is easier to obtain a strong “climatological correlation” than a high “weather correlation.” In that case, considering the day-of-year cycle, a simple linear correlation between two sounding-derived indices and the 30-day moving-average probability of having at least one hailpad hit by hail gave a correlation as high as R = 0.97. On the contrary, considering the next 6 h on a day-to-day forecast, the best linear correlation between the number of hit hailpads in 6 h and an index derived from proximity soundings was only R = −0.36. In other words, a high correlation on highly smoothed climatological data (e.g., mean value for the day of year) does not guarantee a similarly good performance for day-to-day weather forecasts. This is an indication that the processes involved in storm/hail/lightning production are much more complex than how they can be described by the climatological relationships with favorable environments. Similar issues are found also in other meteorological branches, like in dynamic modeling, where Lucarini et al. (2007) have shown that a theory of a dynamical system cannot be derived from the mean fields describing the average state of a flow.
Instead of studying the trends of proxies of favorable environments, one can run for a long period a convection-resolving model (e.g., Leutwyler et al. 2016), to estimate the climatology of storms directly from the simulated convective events. One example is the work by Kahraman et al. (2024), which found that hail basically did not have any trend in central Europe between 1999 and 2018, in contrast with the crowdsourced reports of severe hail of the European Severe Weather Database (ESWD; https://eswd.eu). Another approach (computationally cheaper) could be the event-based “pseudoglobal warming” methodology, like that applied by Mallinson et al. (2023) to simulate hailstorms under climate change. They found that, while CAPE and precipitable water increased in a warmer environment, the hail size during the convective season decreased. However, the approach followed in this work is different, since we will use observations instead of simulations. In particular, we will focus on data observed with a high spatial and temporal resolution on a very small domain, basically the Friuli Venezia Giulia (FVG; NE Italy) region. Many high-quality data, including 31 years of high vertical resolution soundings, a network of more than hundred rain gauges, and a network of about 360 hailpads operated in the FVG plain for 29 years, will be analyzed. Furthermore, the EUCLID CG lightning data, illustrated in Manzato et al. (2022a), will also be used. Thus, we will rely only on real observations, while no data from numerical weather prediction models or reanalysis will be used, apart from a comparison with some ERA5-derived parameters in sections 3c and 4e.
The paper is organized as follows: Section 2 describes in detail the various datasets used; section 3 illustrates the analysis of the sounding-derived parameters, in terms of trends in basic atmospheric variables and in thermodynamic and kinetic indices. Then, in section 4, we present the analysis of trends for all the different storm-related observational datasets. Last, section 5 summarizes the conclusions.
2. Data and methods
In this section, all the different observed datasets used in this work will be shortly presented.
a. Udine radiosoundings at high vertical resolution
The climatic changes in the troposphere above NE Italy will be studied through the observations made by radiosoundings. Figure 1 shows the FVG region, which has a geographical conformation favorable to severe events (Davolio et al. 2016; Miglietta et al. 2016; Manzato et al. 2022a), with mountains up to ∼2700 m high to the north, and the Marano and Grado lagoons, followed by the Adriatic Sea, to the south. In the center of the FVG plain, the Udine radiosounding (“RDS” label in Fig. 1) is performed by Aeronautica Militare (Italian Air Force). Until 17 January 2016, it was released from Campoformido [46.04°N, 13.19°E, WMO identifier (ID) 16044], while, after that day, it was moved 15 km far away, to Rivolto (45.97°N, 13.05°E, WMO ID 16045).
The domain (FVG, NE Italy) where the Udine radiosounding (RDS label) and the 104 rain gauges are located.
Citation: Journal of Applied Meteorology and Climatology 64, 3; 10.1175/JAMC-D-24-0034.1
Since August 2021, the sounding is automatically performed by a Vaisala “autosonde” system, while before it was operated manually. The soundings were done using Vaisala sondes: RS80 from 1992 until February 2002, RS90 until June 2005, RS92 up to November 2016, and then the Vaisala RS41 sonde have been used. While the rising velocity of the balloon tends to be always the same (about 4.4 m s−1), the different sondes improved the sampling frequency of the raw data: The old RS80 saved only one level every 10 s of ascent; the RS90 saved the data in a new vertical level every 2 s, while the RS92 and RS41 save one level every second. For comparison, in troposphere, the raw high-resolution sounding made by RS80 may have ∼250 levels against ∼30 saved in the WMO TEMP format disseminated in the Global Telecommunication System (GTS) circuit, while that made with RS92 may have ∼2500 levels against about 50 of the TEMP format. Manzato (2008) has shown a large sensitivity of indices on the vertical resolution of the radiosounding, particularly for the instability indices heavily depending on the lowest levels. Consequently, in this work, only the original high vertical resolution profiles will be used, not those summarized in the TEMP format.
Udine radiosoundings have been performed with variable frequencies during the years. From 1986 to 2004, four soundings per day were usually performed (0000, 0600, 1200, and 1800 UTC). Since 2004, the 1800 and 0600 UTC soundings have been discontinued in some periods for economic reasons. At present time, only 0000 and 1200 UTC soundings are performed every day, while 0600 and 1800 UTC soundings are occasionally performed in case of meteorological warnings. In this work, only the soundings from April to September 1992–2022 will be studied, as representative of the convective season in FVG (Manzato 2007, 2012). In this period, there are 5602 high-resolution soundings made at 0000 UTC, 2782 at 0600 UTC, 5636 at 1200 UTC, and 2367 at 1800 UTC. Due to the much lower number at 0600 and 1800 UTC, we will present the climatological trends based only on values derived from the 0000 and 1200 UTC soundings. All these 11 238 soundings will be used, even in the presence of cloudy layers or completely saturated profiles.
Every sounding has been analyzed with the SOUND_ANALYS.PY program (Manzato and Morgan 2003), but only values computed with the “Tv method” (virtual temperature method, i.e., using the virtual correction and considering the water vapor content also in the “dry” pseudoadiabat) will be considered. Apart from all the indices computed by this program, also some direct environmental parameters, like temperature T, dewpoint temperature Td, geopotential height z, relative humidity rh, mixing ratio q, normal and saturated vapor partial pressure e and esat, and the equivalent potential temperature Θe [computed with the Bolton (1980) equation] at some mandatory levels (viz., 925, 850, 700, 500, and 300 hPa), will be studied. The data of relative humidity and mixing ratio at 300 hPa are not considered as reliable, because generally the water at such level becomes saturated with respect to ice, while the sounding data seem to provide relative humidity with respect to liquid water. Furthermore, the well-known dry bias of some Vaisala humidity sensors at high elevations (Miloshevich et al. 2006, 2009; Vömel et al. 2007; Moradi et al. 2013) must be considered. For this reason, it may happen that, even if the sonde is inside a (icy) cloud, the relative humidity does not indicate values near saturation.1
List of correlations (with their statistical significance) and trendZ for 52 sounding-derived parameters with statistically significant trends (R > 0.2 and p value < 0.01). All the available soundings at 0000 and 1200 UTC during April–September 1992–2022 (31 years) have been used. MUP means the most unstable parcel. “MM2003” stands for Manzato and Morgan (2003).
As in Table 1, but for the 18 sounding-derived parameters with low or no trend at all (R ≤ 0.2 and p value ≥ 0.01).
b. Two databases of FVG surface stations
One of the typical observations associated with thunderstorms is convective rain. In this work, we do not try to isolate convective rain events from the nonconvective ones, but limit our analysis to the “convective season,” which—in this area—is mainly April–September (e.g., Manzato 2007), with the assumption that the majority of rainfall during that period is convective. The studied domain also includes one of the areas (Julian Prealps) with the highest annual precipitation in Italy (Pavan et al. 2019) and in central Europe (Isotta et al. 2014). The FVG territory is monitored by more than 200 rain gauges, built by two different manufacturers: SIAP-MICROS (https://www.siapmicros.com) and CAE (https://www.cae.it). Most of the rain gauges have a sampling time between 1 and 30 min. However, only 104 stations are selected from the regional mesonet, because these have been considered to comply with the WMO standard (WMO No. 8, 2008) by internal surveillance (Micheletti and Salvador 2014).2 These stations are located at very different elevations, going from the coast up to 1743 m above mean sea level (AMSL). From these data, the mean value of all the 104 station rainfall observations and also the maximum value of all the 104 station rainfall, accumulated in 6 h (starting at 0000, 0600, 1200, and 1800 UTC), 12, and 24 h, have been extracted for the months of April through September from 2006 to 2022. However, the number of “active” stations changed in time, with a very rapid increase in active rain gauges after 2010, as shown in Fig. S1 in the online supplemental material.
Since the number of hourly data to be manually checked for 104 stations is too big, a subselection of 26 of them has been made, choosing high-quality stations among those having the longest lifetime. In particular, the 26 stations written in capital letters in Fig. 1 were all active since 2001, apart from two of them, which started working during that year. For these 26 stations, beyond the automatic controls, also manual quality controls were made, increasing the quality of these 1-h rainfall data.3 Last, from these 26 stations, the mean and maximum rainfall accumulated in 1, 3, and 6 h have been extracted for the months of April through September from 2006 to 2022. Hence, in this work, two different datasets of mean and maximum rainfall in FVG will be studied.
c. Hailpad data in the FVG plain
In the plain of FVG, a hailpad network is operated since 1988. This network relies on the voluntary service of human observers and is maintained and coordinated by the regional meteorological office [Osservatorio Meteorologico Regionale (OSMER)] of Agenzia Regionale per la Protezione dell’Ambiente (ARPA) Friuli Venezia Giulia. It covers an area of about 4500 km2, with a distance between each hailpad station of about 3–4 km on average. A thorough description of the network and of the dataset collected is given in previous works (Morgan 1992; Giaiotti et al. 2003; Manzato 2012; Manzato et al. 2022b).
Hailpad networks represent a cost-effective method to observe hail at ground. The main weaknesses are as follows: 1) volunteers are not always accurate and timely in reporting hail occurrence, leading to potential inhomogeneities and missing events; 2) even if the network is quite dense, it is almost impossible to grasp all the features of hail streaks, because of their much localized nature (Morgan and Towery 1975). Other ground-based hail detection methods have nevertheless their own limitations. For example, crowdsourced hail reports, such as those included in the ESWD, can suffer from both spatial and temporal inhomogeneities, due to different population densities and diurnal efficiency (Taszarek et al. 2020). Networks that consist of more advanced, automatic instruments (e.g., hail impact sensors like those discussed in Kopp et al. 2023) present much higher costs and, at present time, still do not cover wide areas.
Hailpads in FVG plain are deployed from 1 April until 30 September every year. The semiautomatic methodology adopted to analyze the hailpads is described in detail in previous works (Giaiotti et al. 2001; Manzato et al. 2022b), where the reader is referenced. From the analysis, it is possible to retrieve distinctive quantities related to hail characteristics, such as the median diameter of the hailstone distribution per hailpad, D50. In this work, we will consider only the trends of the “annual” (meaning April–September) number of hailpads impacted by hail and of the mean value of D50 during 1988–2016.
d. Calculated convective activity in the FVG plain
In Manzato (2003), a new index to estimate the “6 h convective activity strength” of thunderstorms in the FVG plain was proposed: It mixes the “normalized” logarithmic number of CG lightning [from Centro Elettrotecnico Sperimentale Italiano (CESI)/EUCLID network] together with the maximum 5-min wind speed and the “total rain.” The maximum wind gusts and accumulated rainfall are taken from a subset of 15 among all the stations of the regional network. It is called “CALCA6h” (Calculated Convective Activity in 6 h), and its value is set to zero if there are no CG lightning strikes during the 6-h period. It is available in OSMER since 1995, because before that year no lightning data were available.
Since 2003, CALCA6h has been used operationally in OSMER–ARPA FVG, and experience shows that when it is larger than 0.7 there are some significant thunderstorms, while when it is larger than 0.8 there are usually strong thunderstorms in the FVG plain, almost always reported by local media because of their damages. In this work, we will study not only the annual trend of the mean value of CALCA6h but also the trend of the probability of having CALCA6h larger than 0, 0.7 (significant storm), or 0.8 (severe storms) during 1995–2022.
e. Lightning flashes in the greater Alpine region
All the observations presented in the previous subsections refer to data inside the relatively small FVG region (rain) or only in the FVG plain (hailpads and CALCA6h). However, it would be valuable to extend the analysis also to a larger domain, possibly still connected with the instability trends of the Udine soundings. To study storm climatology in a larger domain, also the EUCLID dataset of CG lightnings studied by Manzato et al. (2022a) will be used. The domain ranges from 42° to 49°N and from 4° to 19°E, including the whole alpine range and portions of the neighboring countries, and can be identified with the GAR. The EUCLID network uses state-of-the-art lightning sensors, such as Vaisala LS7002 sensors, but more technical details can be found in Manzato et al. (2022a).
Figure 2 shows the climatology of the EUCLID lightning density (LD) evaluated on the GAR domain during 2005–19 (adapted from Manzato et al. 2022a). It is possible to appreciate how NE Italy and the Julian Prealps are among the regions with the highest peak of CG flashes in the domain. We will study the trends of the GAR mean LD, considered both as mean “annual intensity” (mean value of number of CG in 10 min for each April–September period) and as mean “annual frequency” (probability of having a number of CG in 10 min larger than some given thresholds, during April–September of each year).
The EUCLID CG lightning climatology evaluated on the “GAR” domain (adapted from Manzato et al. 2022a).
Citation: Journal of Applied Meteorology and Climatology 64, 3; 10.1175/JAMC-D-24-0034.1
3. Trends of sounding-derived parameters
In this section, all the “annual” (meaning April–September) trends of the 70 parameters listed in Tables 1 and 2 will be discussed. In these tables, the parameters are listed in order of decreasing correlation R2 and trendZ, with Table 2 showing those having very low or no trend at all.
a. Trend of atmospheric variables at some mandatory levels
First, we will discuss the results of six environmental variables (viz., T, z, Td, q, Θe, and rh) at the mandatory levels of 925, 850, 700, 500, and 300 hPa. Figure 3 presents their annual mean anomaly (except for rh, whose original values are shown) and the linear fit used to compute the trend, i.e., the linear slope in 10 yr of the normalized values. Except for rh, all the other correlations with years are robust and statistically significant, with relatively high R2 and low p value (the maximum p value is 1.6 × 10−3 for z at 925 hPa).
Trends of the anomalies for temperature, geopotential height, dewpoint temperature, mixing ratio, and equivalent potential temperature Θe at some mandatory levels (925, 850, 700, 500, and 300 hPa). The last figure is made for relative humidity, not its anomaly. Note that rh and q at 300 hPa are not shown. All available 1200 and 0000 UTC soundings during April–September 1992–2022 are used.
Citation: Journal of Applied Meteorology and Climatology 64, 3; 10.1175/JAMC-D-24-0034.1
The first figure (top left) shows the anomaly of T and its fitted trend. Temperatures during 1992–2022 have a strongly positive trend. At first glance, the anomalies at different levels look similar, with an average increase of about 1.63°C in 31 years [0.53°C (10 yr)−1], as will be shown in Table A1. Giorgi et al. (1997) and Auer et al. (2007) found an elevation-dependent warming effect, with more warming for surface stations located at higher elevations than for those at lower elevations. However, in the Trentino region [northern (N) Italy], Tudoroiu et al. (2016) found opposite results, i.e., a negative elevation-dependent effect, with more elevated stations warming less than those at lower altitudes.
This discussion is out of the main scope of this work, because sounding data are not surface station data, but we found the lowest correlation and slope [R2 = 0.34 and trendZ = 0.66 (10 yr)−1] for the level at 925 hPa, while the highest ones [R2 = 0.52 and trendZ = 0.82 (10 yr)−1] are at 500- and 300-hPa levels. The difference is not very large, but, from these sounding data in NE Italy, the higher part of the troposphere seems to warm at a higher rate than the lower part of the troposphere, similar to what was found by Giorgi et al. (1997) and Auer et al. (2007) for mountain’s stations with respect to valley/plain stations.
Looking at the geopotential height z, one can see very similar behavior, but with more difference in trends between the lowest level [R2 = 0.30; trendZ = 0.63 (10 yr)−1 at 925 hPa] and the highest one [R2 = 0.56; trendZ = 0.85 (10 yr)−1 at 300 hPa]. This difference suggests that the increase in geopotential at 300 hPa (higher trendZ than for T) and the relatively smaller increase at 925 hPa (lower trendZ than for T) are not completely explained by the thermal component, but probably there could be also a dynamical effect. In any case, also the geopotential above NE Italy has a significant positive trend in 1992–2022, mostly explained directly by global warming (same trendZ as T at 850, 700, and 500 hPa).
Correlations and trendZ are even higher for Td at 925, 850, and 300 hPa, on the order of R2 = 0.62 and trendZ = 0.90 (10 yr)−1, respectively. Instead, at 700 hPa, Td has similar metrics to T. The term Td at 500 hPa has a slightly lower correlation and trendZ with respect to the other levels.
Also, Θe has high correlations (R2 = 0.63 or 0.64) and trendZ values [0.90 (10 yr)−1 or 0.91 (10 yr)−1] at 925, 850, and 700 hPa, while the increase is slightly smaller at 500 and 300 hPa. In any case, Θe has—in general—the second highest correlation and trends after q, while temperature has the lowest metrics of Fig. 3 (excluding rh). However, it is interesting to note that, while the temperature anomaly increases more at 500 hPa than at 925 hPa, for the Θe anomaly the opposite is true, because at lower levels there is a much larger increase in moisture. Note that increasing Θe in the lowest levels tends to raise the potential instability, while increasing T (hence also the saturated Θe, i.e., Θes) at 500 hPa tends to decrease it; thus, no straight conclusion on the trend of instability can be made from that.
The mixing ratio at 925, 850, and 700 hPa has the highest correlation (in the 0.65–0.70 range) and trendZ [in the range between 0.92 (10 yr)−1 and 0.95 (10 yr)−1] of all these mandatory level variables. On the other hand, q at 500 hPa seems to increase less [trendZ = 0.76 (10 yr)−1], but that could be due to the underestimation of rh in the midtroposphere in the sounding data. Note that, being water vapor the most important natural greenhouse gas, this large increase in absolute moisture in the troposphere will provide a strong positive feedback to climate warming (Bony et al. 2006).
We made an additional study to check if the different radiosonde types produced some systematic effect in the moisture measurements, possibly related to the radiative effects during daytime (Vömel et al. 2007; Miloshevich et al. 2009). In particular, Fig. 4 shows the annual trend of the ratio between q during nighttime (0000 UTC) and q during daytime (1200 UTC). While—obviously—rh during night is usually higher than during day, the mean absolute moisture at a fixed level is not expected to change in 12 h, in particular in the free troposphere and when considering values averaged over 6 months (April–September). Surprisingly, the ratio q00/q12 is systematically larger than one, except at 925 hPa after 2016 (when the Vaisala RS-41 replaced the RS-92). Moreover, this ratio is the largest at 700 hPa, followed by the one at 500 hPa.
Trends of the ratio between the mixing ratios observed during night (0000 UTC) and during daylight (1200 UTC) at the mandatory levels of 925, 850, 700, and 500 hPa, with their linear fit. Only soundings during April–September 1992–2022 are used.
Citation: Journal of Applied Meteorology and Climatology 64, 3; 10.1175/JAMC-D-24-0034.1
To explain this behavior, we can only suspect that there could be an observational error, due to the strong radiative effect on the rh sensor. In particular, the radiative effect at 1200 UTC is stronger in the free atmosphere than at the lowest levels, because the atmosphere transmissivity is higher than inside the PBL and because there is less shielding offered by the low-level clouds (like the stratocumulus forming at the PBL top). While the value of q00/q12 is the lowest at 925 hPa (with a reversal of magnitudes, i.e., q12 > q00, after 2016), its trend is the highest (R = 0.29 with p value = 0.002), followed by the trend of ratios at 850 and 700 hPa. Except for the large spread visible at 500 hPa (likely because the absolute values of q are very small), all the other trends follow quite well their linear fit. This suggests that there are no tipping points related to the use of different radiosondes in different periods. In conclusion, even if there could be some error in the absolute value of moisture during daytime (with q12 about 10% smaller than q00), it does not seem that this issue could influence the trends found in our dataset for the moisture-related indices.
Last, Fig. 3 shows the rh trend, which is the lowest among all these atmospheric variables. For this quantity, we show the absolute values instead of anomalies, because it is interesting to note how close the absolute rh values are for the 925-, 850-, and 700-hPa levels: In fact, all of them have average values in the 59%–69% range (while at 500 hPa, the mean rh value is only around 40%). Also, trends at both 700 and 500 hPa are extremely small. Thus, it seems that the atmosphere reacts to global warming by trying to conserve relative humidity as much as possible, in particular in the lower troposphere, despite the large increase in q. This topic deserves a deeper investigation that will be done in the appendix. Here, we just note that small or not significant trends in rh, eventually accompanied by positive trends in q, were found by previous works (e.g., Dai 2006; McCarthy et al. 2009).
b. Trends of other sounding-derived indices
After the analysis of basic atmospheric variables at some standard pressure levels, we now turn to the assessment of sounding-derived, thermodynamic, and kinetic indices. Figures 5 and 6 show the trend of the mean value and of percentiles for some of the indices listed in Table 1, i.e., those having the largest trends during April–September 1992–2022. Looking at Table 1, it is possible to see that the parameter having the highest correlation R2 and trendZ [0.73 and 0.94 (10 yr)−1 for its mean value, shown in Fig. 5 together with the 10th and 90th percentiles] is the precipitable water (PWE). It represents the integrated moisture along the sounding profile (also called column-integrated water vapor), from the surface up to 12 km. In Fig. 5, it is possible to see that PWE changed from an average of about 23 mm in 1992 to about 28 mm in 2022, which is a relative increase of 21%. Positive trends of PWE in the most recent period were already found in previous studies by Ross and Elliott (2001), Durre et al. (2009), and Mattar et al. (2011). However, the absolute trend of about 1.57 mm per decade found in Udine is about 3 times higher than those found in other studies [e.g., the values in Table 1 of Sherwood et al. (2010) or the 0.4 mm (10 yr)−1 found by Mattar et al. (2011) for the Alps during 1973–2003]. Such a large increase in PWE means that there is a higher probability of rain in the troposphere. In fact, Yano and Manzato (2022) have shown that there is no direct relationship between PWE and the quantity of rain at the surface. In particular, analyzing the PWE data from Udine soundings and rain data from some FVG stations, they found that heavy rain events have about the same probability when PWE is sufficiently large (e.g., PWE ≥ 30 mm).
Trends of eight indices derived from the Udine–Campoformido sounding during 31 years (1992–2022), including the linear fit of the mean value and of the 10th or 90th percentile. Only April–September 0000 and 1200 UTC soundings are used.
Citation: Journal of Applied Meteorology and Climatology 64, 3; 10.1175/JAMC-D-24-0034.1
As in Fig. 5, but for the other seven sounding-derived indices with significant trends. (bottom right) For CIN, it is shown also the trend for the subset of cases with significant instability (CAPE > 500J kg−1).
Citation: Journal of Applied Meteorology and Climatology 64, 3; 10.1175/JAMC-D-24-0034.1
Similarly to PWE, also the mixing ratio of the most unstable parcel (Mix) shows a very high correlation and trendZ, as found for q at 925 hPa in Fig. 3. Mix is also well correlated with the temperature at the lifting condensation level (Tbase), estimating the cloud-base temperature. Figure 5 clearly shows the increase in Mix and Tbase in the last 31 years. In particular, in recent years, the mean value of temperature of the cloud base in April–September has become almost as high as +10°C, due to a large absolute trend of about 0.94°C (10 yr)−1. Note that a higher cloud-base temperature should favor “warm rain” against “cold rain” formation through mixed-phase (Wegener–Bergeron–Findeisen) processes.
Two other variables with high correlations and trendZ values are the shear variables, defined as the normalized length of the hodograph
Lastly, Fig. 5 shows also the significant trend of the severe weather threat (SWEAT) index (mixing potential instability and wind shear together), of the most unstable parcel (MUP) Θe, and of the updraft velocity (UpDr). In Sound_Analys.PY, UpDr is computed by converting the buoyancy energy into the corresponding vertical velocity (with the usual equation
Figure 6 shows the trend of four commonly used estimates of potential instability, namely, CAPE, maximum buoyancy (MaxBuo) [i.e., Max(Θe)low − Min(Θes)mid], most unstable lifted index (MULI), and EHI. All of them have a very significant trend, with a correlation ranging from 0.50 to 0.59 and a ‖trendZ‖ = ranging from 0.78 (10 yr)−1 to 0.84 (10 yr)−1. These metrics are higher than what is found for T in most of the mandatory levels, due to the associated large trend also in moisture. Hence, in these 31 years, there has been a very large increase in potential instability above NE Italy, which is highly statistically significant (p value < 10−5). Figure 6 shows that also the environmental height of wet bulb at the 0°C level (WBZ) is rising, confirming what was found for Tbase. In general, a higher freezing level should favor the melting of hail, hence decreasing the probability of reaching the ground for the smallest hailstones.
The last row of Fig. 6 shows the CIN, with higher values (lower in absolute value, since the inhibition energy is considered negative) indicating less convective inhibition. The left figure shows CIN from all the April–September 1992–2022 soundings and highlights a relatively large increasing trend, i.e., less inhibition in more recent years, differently from what was suggested by Trapp and Hoogewind (2016) and Chen et al. (2020). Both the increase in potential instability and the decrease in CIN (as absolute value) should promote the development of thunderstorms. Last, the right side of the last row shows the trend of the Udine CIN computed only for cases with significant potential instability (those having CAPE > 500 J kg−1, but similar results are found for CAPE > 100 or CAPE > 1000 J kg−1). In this subsample, the convection inhibition has a slightly increasing trend [trendZ = 0.17 (10 yr)−1], which is not statistically significant (p value 0.4). Thus, looking only at the significantly unstable cases, it is possible to say that CIN has no trend. In any case, this analysis does not show that, in NE Italy, there are less favorable environments for CI in recent years compared to the past. On the contrary, the environment may be slightly more favorable due to more instability and less convective inhibition.
c. Comparison with some ERA5 indices
Despite the fact that this work focuses on observed data, it is interesting to compare some of our results with what is simulated by the widely used ERA5 reanalysis. In particular, we will compare PWE, that is, the index showing the highest trend, with the corresponding total water content provided by ERA5 above Udine, and we will compare CAPE and CIN indices with the most unstable CAPE and CIN provided by ERA5.
Figure 7 shows that also the total column water vapor content seen by ERA5 increases in time, even if with a much lower slope than the sounding-derived PWE. In fact, its absolute trend is only 0.74 mm (10 yr)−1 and R2 is only 0.41. This difference is mainly due to the fact that, before 2001, the mean water vapor content above Udine in ERA5 was always higher than that observed by the Udine sounding (which should be assimilated in ERA5), while, after 2014, it is always lower. These differences seem quite systematic and highlight how studies based on ERA5 reanalysis can produce substantially different results from those based on observed soundings, as found also by Taszarek et al. (2021).
Comparison of the trends of PWE, CAPE, and CIN derived by Udine soundings and by the ERA5 reanalysis above the 46.0°N, 13.0°E grid point. Note that, given that sounding-derived indices may have some missing cases, the number of cases N for ERA5 is higher than that for RDS.
Citation: Journal of Applied Meteorology and Climatology 64, 3; 10.1175/JAMC-D-24-0034.1
In the case of CAPE, the trends are more similar, even if ERA5 has again a lower slope [62 J kg−1 (10 yr)−1 vs 103 J kg−1 (10 yr)−1], likely because it has a systematically lower value. On the contrary, ERA5-derived CIN above Udine has an opposite slope with respect to the sounding-derived one. While the CIN from sounding is becoming smaller year after year (i.e., atmosphere is more unstable), at a rate of 44 J kg−1 every 10 years, CIN from ERA5 is becoming more “stable” [as was suggested by Trapp and Hoogewind (2016) and by Chen et al. (2020)], at a rate of −12 J kg−1 (10 yr)−1. While the different absolute values of CAPE and CIN from soundings and ERA5 could be due to different methods used to compute them in Sound_Analys.PY and by ECMWF, the opposite slope found for CIN is quite surprising and should deserve further investigation.
d. How do the distributions change with time?
In section 3b, we have analyzed the trends of the indices’ annual values. By looking at Figs. 5 and 6, we have shown that they indicate increasing instability and moisture content. When the annual mean value of a parameter increases, it can result from either 1) the entire distribution is shifted, mainly preserving its shape, with most of the change coming from a relatively small shift of the majority, close to the distribution mode; or 2) the distribution changes its shape, with the tail of the—relatively few—extreme values substantially increasing. Thus, the distribution change may or may not involve a change in “symmetry,” which can be approximated by its skewness.
For example, an increase in time of the mean value of CAPE can be due to more cases with CAPE > 0 with respect to the past, or to the fact that there is a similar number of unstable cases, but with a higher frequency of extreme values. Last, both effects can act simultaneously. To give a visual perspective of how some index distributions may evolve with time, Fig. 8 shows the distributions for the subset of the 1992–97 (solid) and 2017–22 (dashed) periods. The three indices shown (PWE, CAPE, and cloud depth) are characterized by three different kinds of distributions: bell-shaped, exponential, and bimodal, respectively. For PWE, most of the change comes from the central values, while for CAPE, there are both a decrease in the null cases and an increase in the right tail.
Comparison for distributions of PWE, CAPE, and EL-LCL, derived from the Udine soundings of April–September 0000 and 1200 UTC, during the first six (solid line) and the last six (dashed) years of the studied period. Since the total number of cases N analyzed is slightly different in the two periods, it is explicitly reported.
Citation: Journal of Applied Meteorology and Climatology 64, 3; 10.1175/JAMC-D-24-0034.1
Figures 5 and 6 only show the trends for the mean and for the percentiles of the parameter’s intensity for any given year but do not give information about the change in their shape. A different approach is obtained by setting some thresholds on the interannual CAPE distribution and looking at how the probability of exceeding those thresholds changes in time. Figure 9 shows this kind of frequency analysis, for some high percentiles (75%, 90%, and 95%) of the CAPE (above) and PWE (below) entire distributions. It is possible to see that their R2 and trendZ are similar—but lower—to those found for the corresponding mean intensity in Table 1. In particular, for CAPE [but the same trend is found also for other instability indices like MULI and Showalter index (ShowI)], both correlation R2 and trendZ monotonically decrease when increasing the percentile, suggesting that there is no higher increase in the extreme values with respect to the rest of the tail. For example, the trend for 95% is lower than that for the 75th percentile. It can be interesting to study also the trend of the number of potentially unstable cases, defined as CAPE > 0, which are as many as 82.5% of the total, because during April–September the troposphere is very often potentially unstable above NE Italy. Also, this subset of cases (corresponding to the 17.5th percentile) has a positive trend, even if with a lower slope than those found for the other percentiles of Fig. 9. For PWE, the highest trend in frequency is found for the 90th percentile, and also in this case, R2 and trendZ are lower than what is found for the mean intensity of Table 1.
Trends of (top) CAPE and (bottom) PWE above the 75th, 90th, and 95th percentiles of their distribution. The trend for the frequency of annual cases with CAPE > 0 J kg−1 is also shown (note that 12.5% of cases have CAPE = 0 J kg−1, on average).
Citation: Journal of Applied Meteorology and Climatology 64, 3; 10.1175/JAMC-D-24-0034.1
As explained above, an additional perspective on what happens to the shape of a distribution can be gained by studying the trend of its skewness. The skewness, or third moment, measures the degree of symmetry: An almost Gaussian distribution has an almost null skewness (like PWE during April–September), while a zero-bounded distribution has a high positive skewness (e.g., CAPE, with a value of 2.3). We studied the trends of the annual skewness for all the 70 indices listed in Tables 1 and 2. The skewness of most indices has a low trend and low correlation with time, meaning that the distribution does not change its shape, but mostly shifts “rigidly.” Table 3 shows only 21 parameters having a not negligible linear trend for their skewness, i.e., R2 > 0.1 with p value < 0.1.
The linear fit of the 21 parameters having a not negligible trend in their annual skewness. The first column is the mean skewness value during April–September 1992–2022. The other three columns show the slope, correlation, and statistical significance of the linear regression with respect to years.
Of all the parameters shown in Table 3, EHI has the highest skewness, while PWE has the lowest one. The parameters with the largest skewness trends are the PW inside the “cloud” (PWC; computed by integrating q of the lifted parcel from LCL up to EL, which is defined only if there is a LFC, i.e., MaxBuo > 0), CAPE, UpDr, CIN, and the cloud depth (EL-LCL). The first seven variables, except for EL-LCL, have a zero-bounded distribution; hence, they have an “intrinsic” high skewness. However, EL-LCL, LFC, and meridional water vapor flux (VFlux) have a more “bell-shaped” distribution, but still have a significant trend of their skewness, with 0.21 ≤ R2 ≤ 0.38 and p value < 0.01. As shown in Table 1, EL-LCL has a strong increasing trend [LCL tends to become lower, with an absolute trend of −56 m (10 yr)−1, while the cloud top tends to become higher than in the past, with a trend of 443 m (10 yr)−1]. However, Table 3 shows that its skewness is constantly decreasing, passing from positive values before 2002 to negative values after then (not shown). This means that the left tail (values below 3 km) decreases, while the right tail (values above 9 km) increases. While for PWC, CAPE, and UpDr, a negative slope of the skewness fit means that their distribution tends to become more symmetric (e.g., CAPE tends to have fewer null cases in recent years), for CIN (negative energy that must be provided to the parcel), a negative slope means that its distribution is becoming more asymmetric with time. In fact, if we study the trend of the cases having CIN ≥ −10 J kg−1, which represents the 25th percentile, they are increasing in time [R2 = 0.10, p value = 0.08, and trendZ = 0.35 (10 yr)−1].
It is also interesting to note that the parameter having the largest mean value trend, which is the precipitable water in the whole tropospheric column (PWE), does not have a substantial trend in its skewness, and hence, its distribution is basically shifting to the right without changing much its shape, as shown by Fig. 8. On the other side, the parameter changing the most in its shape is the PWC, because LCL is decreasing [trendZ = −0.50 (10 yr)−1], while the equilibrium level is increasing [trendZ = 0.70 (10 yr)−1] and hence q is integrated on a higher column.
To show an example, some of the skewness trends are shown in Fig. S2 (for PWC, CAPE, UpDr, CIN, VFlux, and PWE). Trends toward more asymmetric distributions are only found for these indices: CIN, LFC, rh at 850 hPa, e and q at 700 hPa, PBL, Td at 500 and 850 hPa, LLWv, and PWE, which have the same sign for the mean skewness and for the slope of their fit. However, the skewness value and its trend should be interpreted differently for “bounded” or bell-shaped distributions, with the same change in skewness being more significant for the nonbounded variables.
4. Trends of storm-related observations
In the previous section, we analyzed the trends of 31 years of sounding-derived parameters. These trends highlighted, against any doubt, that the atmosphere in NE Italy is warming and, in particular, “moistening.” In fact, the parameter that has shown the most significant increase is PWE. This evidence could induce to expect an increase in rain or at least in the frequency of rainy events. While some works proposed that rain trend scales even more than the 7% °C−1 of the water-holding capacity, due to the Clausius–Clapeyron equation, others (like IPCC 2001; Allen and Ingram 2002; Skliris et al. 2016) state that some energy budget constraints would limit the climatological rain trend to lower rates.
For example, Ciccarelli et al. (2008) found no substantial rain trend in western Italy from surface observations, while model simulations by Pall et al. (2007) and O’Gorman and Schneider (2009) showed that the rain increase in midlatitudes is lower than the rate given by Clausius–Clapeyron law. On the contrary, climate simulations by Vergara-Temprado et al. (2021) found that the intensities of subhourly precipitation tend to scale with about 6.5% °C−1 over Europe, while Molnar et al. (2015) found increases in rain in the range of 6.5%–13% °C−1 analyzing 30 years of rain from 59 stations in Switzerland. Berg et al. (2013) found an increase in convective precipitation in Germany up to 14% °C−1. In fact, the rate of moisture and rain increase with global warming is highly debated in many works (among others, Trenberth et al. 2003; Westra et al. 2014; Ban et al. 2015; Martinkova and Kysely 2020; Neelin et al. 2022). Here, we will contribute to this discussion by reporting what is found in the data available from the rain gauge network in FVG.
Moreover, from Table 1, it is also evident a clear increase in potential instability [SWEAT, MUP ThetaE, CAPE, MaxBuo, difference in temperature at −15°C cloud level (DTC), and MULI], as well as a corresponding decrease in CIN. More favorable environments would induce to expect an increase in the frequency of severe thunderstorm events, including more hail and more lightning production. However, one of the questions investigated in this work is if the trends of few sounding-derived parameters (like instability, CIN, and shear) are sufficient to estimate the real trend of thunderstorms or if there may be also other influencing factors.
For example, Sherwood et al. (2010) described the mechanisms under which, in a much more moistened midtroposphere, the evaporation of the rain falling from the cloud would be reduced, leading to weaker downdrafts. Downdrafts are a key factor for the internal dynamic of severe storms, like supercells (e.g., for their interplay with updraft to create low-level rotation), and for the production of significant cold pools that often are able to trigger “secondary” convection (Purdom 1976; Tompkins 2001; Torri et al. 2015; Hirt et al. 2020). Moreover, a strong downdraft is essential to produce strong vorticity near the ground, needed for the genesis of severe tornadoes (Davies-Jones 2015), even if excessively strong downdrafts have been found to be detrimental to tornado formation by Markowski and Richardson (2009). Thus, weaker downdrafts should lead to weaker supercells, fewer storms, and fewer severe tornadoes. Betts and Silva Dias (1979) wrote that strong downdrafts can only occur if the lapse rate is close to the dry-adiabatic one, i.e., if the atmosphere is far from saturation. The moistening found in this work, in terms of rh and hence distance from saturation, is not very large, since rh has relative increases of only 3% in the low levels, where downdraft usually originated (Torri and Kuang 2016). Then, it is also not so clear if this small increase in rh could lead to a significant decrease in the average strength of downdrafts and hence to a change in the storm trend, as suggested by Sherwood et al. (2010).
Moreover, the trend of the downdraft potential index (Morgan and Tuttle 1984; Manzato 2012), which aims at giving a rough estimation of the downdraft strength, was found to actually increase in Table 1, not to decrease. Thus, the sounding-derived analysis tends to suggest that there should be a higher frequency of strong supercells, with stronger cold pools that should generate more secondary convection, and of severe tornadoes. Of course, a stronger downdraft should also increase the probability of severe linear outflows and microbursts. Also, simulations by Torri and Kuang (2016) showed that downdrafts might be more sensitive to the precipitation rate than to environmental rh, while simulations by Windmiller et al. (2023) showed that precipitation—and hence downdraft characteristics—can be better estimated by the specific updraft characteristics (including area, mass flux, and speed) than by generic environmental properties, like CAPE or rh.4 In conclusion, convection is a much more complex phenomenon than what few sounding-derived parameters can describe. Often, it is not possible to deeply understand what will happen to a highly nonlinear system, like a thunderstorm, simply by looking at the changes in some environmental characteristics. Consequently, it is necessary to directly investigate the observations of storm occurrence and intensity, as in the following.
a. Rainfall in FVG
First, the dataset of the 104 6-h rain stations is studied. Figure S1 shows how many of these 104 stations were active during 2006–22. Figure 10 studies the frequency analysis for the 104 rain gauges accumulated in 6, 12, and 24 h, respectively. The left side shows the average rain of all these stations, while the right side shows the maximum rain among the 104 values per each time interval. Three frequency trends, together with their linear fit, are shown: the probability (as a number of cases) of exceeding a fixed threshold (0.4 mm for MeanRain or 1 mm for MaxRain) or of exceeding the 90th, 95th, or 99th percentiles of the interannual distribution. The statistical significance of these trends is much lower than those found for most of the sounding-derived parameters. The rain trend with the highest significance (p value as low as 0.036) is MaxRain > 1 mm (12 h)−1, which has a positive trend with R2 = 0.26 and trendZ as high as 1.0 (10 yr)−1. It is followed by the same field (MaxRain > 1 mm, as shown by the light gray fit in the right column of Fig. 10) in 6 and 24 h, which have R2 = 0.24, a p value much lower than for all the other cases and a positive trendZ, close to 1 (10 yr)−1. Regarding the highest percentiles (i.e., frequency of heavy rain events), the most significant one (but with p value as high as 0.17) is MeanRain > 12 mm in 12 h that has R2 = 0.12 and a negative slope [trendZ −0.69 (10 yr)−1]. Thus, the most statistically robust conclusion that can be derived from Fig. 10 is that only the rainfall cases which include the weak local events (when a single station reports a MaxRain of at least 1 mm in 6, 12, or 24 h) are increasing with time. However, Fig. S1 shows that during the studied period there has been a significant change in the number of active stations, so the increased number of localized rainfalls could be due to the larger number of sampling stations. In fact, if we remove the data from 2006 to 2009 and repeat the analysis of Fig. 10, the 1-mm positive trends disappear, while some negative trends in the highest percentiles emerge (not shown). In any case, there is no evidence of an increase in the frequency of significant or heavy rain events, possibly the opposite.
Trends of (left) mean and (right) maximum rain in 104 stations of FVG, accumulated in (top) 6, (middle) 12, and (bottom) 24 h, during April–September 2006–22. The probability of exceeding 0.4 (for mean) or 1 mm (for max) accumulated rain in the period (light gray); the 90% (medium gray), 95% (dark gray), and 99% (black) of the rainfall distribution and their linear fit are shown. For 12 and 24 h, the 99th percentile is not shown because there are too few cases.
Citation: Journal of Applied Meteorology and Climatology 64, 3; 10.1175/JAMC-D-24-0034.1
To alleviate the issue due to a changing number of working stations and to investigate flash-flood events with a higher sampling time, Fig. 11 has been made using the hourly data of the 26 stations having a complete record during April–September 2001–22. The probability of exceeding the 99th percentile is particularly interesting to study short-duration intense rain events. However, no rain trend is found in Fig. 11. The most significant trend has a p value as high as 0.35 and is found for 99% of MeanRain in 3 h, which has R2 of only 0.04 and a decreasing trend [trendZ = −0.32 (10 yr)−1]. In conclusion, from both of these frequency analyses of the 104 rain gauges in 2006–02 and of the 26 rain gauges during 2001–22, it can be concluded that there is no statistically robust trend for the rain in FVG, in particular when accumulated during periods from 1 to 12 h, which are typical time scales for storms crossing the FVG area.
As in Fig. 10, but for rain in 26 stations of FVG, accumulated in (top) 1, (middle) 3, and (bottom) 6 h, during April–September 2001–22.
Citation: Journal of Applied Meteorology and Climatology 64, 3; 10.1175/JAMC-D-24-0034.1
If a similar analysis is repeated for the rain intensities (i.e., the average value of the Mean/MaxRain or of some fixed percentiles of the yearly rain distribution), the result is the same: No rain trend is found in FVG, as shown in Fig. S3. Also, if the same intensity analysis is repeated independently for each of the 6 studied months, again no trend is found (not shown). Last, one can think that the results can be highly dependent on the complex FVG orography and in particular on the contribution of stations located in the mountains. However, if we repeat the same intensity analysis based only on plain stations, then the same trends found for the 104 stations are still evident, as shown by Figs. S4 and S5. The same is found also for longer time series, by the comparison of the mean intensity trend derived from all 26 hourly data stations and from a subset of them, including only the 12 stations located in the FVG plain (not shown). From this analysis, it is clear that, during April–September in FVG, there is no evident rain trend, as already found for western Italy by Ciccarelli et al. (2008). Possibly, there could be an increase in the frequency of all the rain events, including cases when also the drizzle events are counted (Fig. 10). If the rain events are more frequent in recent years than in the past, while the probability to exceed the highest percentiles is almost constant, it is not excluded that the total rain can increase. To test this hypothesis, Fig. 12 shows the sum of the rain of all the 26 stations analyzed in Fig. 11, for the period April–September 2002–22 (when all the 26 stations had complete hourly records). This sum of rain values from more stations represents a sort of “areal flux of rain” in FVG. The gray bars and their linear fit shown in Fig. 12 correspond to the total rain per year: Again, there is a negligible correlation with years (R2 = 0.02; p value = 0.50). Instead, the thin and thick black lines show the yearly frequency of cases with hourly total rain above the 90th and 95th percentiles of their total distribution. The highest correlation (R2 = 0.13; p value = 0.10) is found for the 90th percentile, which shows a negative slope. In conclusion, also from the areal flux of rain in FVG, there is no sign of an increase in total convective rain with global warming, despite the large increase in precipitable water and absolute moisture.
The annual “total flux of convective rain” in FVG, derived by summing together the accumulated rain during April–September of each of the 26 stations of the previous figure. The scale on the left refers to the gray bars (total rain), while the scale on the right refers to the two black lines (% of intense cases). Note that 2001 is missing because two stations did not have complete records during that year.
Citation: Journal of Applied Meteorology and Climatology 64, 3; 10.1175/JAMC-D-24-0034.1
b. Hailfall in FVG
In this section, we will discuss the same data analyzed by Manzato et al. (2022b, where the reader can find more details), which are based on the hailpads collected only in the plain of the FVG region, during April–September 1988–2016. Figure 13 shows that the total number of collected hailpads has very big variations (probably also depending on the different dedication of the volunteers, in particular during the last years). Apart from that, there is basically no trend, because the negative slope is not statistically robust: R2 = 0.10 and p value = 0.10.
Trends of the total number (top) of hailpads impacted by hail in FVG plain and (bottom) of the annual mean value of the median hailstone size per hit hailpad during April–September 1988–2016. Adapted from Manzato et al. (2022b).
Citation: Journal of Applied Meteorology and Climatology 64, 3; 10.1175/JAMC-D-24-0034.1
However, Manzato et al. (2022b) studied also some characteristics of the hailstones that impacted each hailpad. In particular, they found a significant increase in time of the average 50th percentile of the hailstone diameter distribution D50 (of each hailpad). As shown by the gray line and fit of Fig. 13, there is a positive trend of this average dimension from about 7.1 mm in 1988 to about 7.6 mm in 2016 [R2 = 0.28 and p value = 0.003, with trendZ = 0.68 (10 yr−1)]. As a comparison with Table 1, similar metrics are found for the downdraft potential index or for the bulk shear BS850, which was found to be a very important predictor in multivariate hail forecasting by Manzato (2013). The increase in D50 also fits well with the increasing trends of HD and WBZ, shown in Table 1 and discussed at the end of section 3b. From this evidence, we can conclude that the frequency of hailstorm occurrence in FVG shows no trend, while hailstones have a significant trend toward larger sizes, on average. These results are compatible with those obtained by model simulations by Trapp et al. (2019) for summer hail in the United States, where they found less hailstorm days but with larger hailstones.
c. Calculated convective activity in FVG
As discussed in section 2d, Manzato (2003) introduced a “multiaspect” estimate of convective severity in the FVG plain, called CALCA6h. Figure 14 shows the trends of CALCA6h during April–September 1995–2022, in two different panels. Figure 14 depicts the mean annual CALCA6h intensity and the likelihood of exceeding the 90th or 95th percentiles. All these trends have a low p value and a negative slope. In particular, the mean value of CALCA6h has a high correlation with years, R2 = 0.40 (p value = 3 × 10−4), and a high value trendZ [−0.77 (10 yr)−1]. It means either that there are less thunderstorms in the more recent years or that they have reduced their severity.
Trends of CALCA6h on the plain of FVG (top) as mean value (black line) and 90th or 95th percentiles of the distribution and (bottom) as percentage of cases exceeding a given threshold with respect to the total number of cases greater than 0 (gray bars), during April–September 1995–2022.
Citation: Journal of Applied Meteorology and Climatology 64, 3; 10.1175/JAMC-D-24-0034.1
To shed more light on these two possibilities, the bottom panel shows, as gray bars and linear fit, a frequency analysis. Gray bars are the time series of the total number of cases with CALCA6h > 0 (i.e., all the 6 h cases that produced at least one CG flash in the FVG plain). It is evident that the number of storms (including the less severe) is decreasing in time, with a correlation as high as R2 = 0.52 (p value = 2 × 10−5) and a high value of trendZ [−0.87 (10 yr)−1]. Hence, the decrease in the mean value of CALCA6h (upper panel) can be explained by a remarkably lower number of storm occurrences in the FVG plain. However, the number of cases with CALCA6h > 0 seems to have a tipping point between 2002 and 2003 that could be due to a different sensitivity of the lightning network to single flash events, as for example, a change in the stroke-to-flash grouping algorithm (e.g., San Segundo et al. 2020). For that reason, also the trends of the number of cases with CALCA6h above higher thresholds have been investigated (not shown), finding that all of them have a negative slope, even if with lower statistical significance. As an example, CALCA6h > 0.6 has a correlation of R2 = 0.19 (p value 0.02) and trendZ = −0.53 (10 yr)−1, while CALCA6h > 0.8 (representing about 99% of the intensity distribution) has R2 = 0.10 (p value 0.11) and trendZ = −0.38 (10 yr)−1.
In order to highlight the number of severe events relative to the total number of storms, the thin and thick lines in the bottom panel of Fig. 14 show the CALCA6h > 0.7 (significant storm) and CALCA6h > 0.8 (severe storms) frequencies relative to the number of cases with CALCA6h > 0. Both of them have a positive trend, particularly high for CALCA6h > 0.7. In practice, the storms in the FVG plain are decreasing, both for the total number of events (CALCA6h > 0) and for the highest percentiles of the CALCA6h distribution. However, the number of significant and severe storms relative to the total number of occurring storms is increasing. This fact probably influences our subjective perception that there are more severe events nowadays. This result is coherent with what has been found for hail: There seem to be less storms, but when they occur, it is more likely that they are severe.
d. Lightning flashes in greater Alpine region
For computing CALCA6h, also the CG lightning is considered, but only in the—relatively small—domain of the FVG plain. Instead, Fig. 2 shows the larger GAR domain and the CG flash climatology derived from the EUCLID dataset studied by Manzato et al. (2022a), where the reader can find more technical details. In this section, we will present some lightning trends, derived from the same domain during 2005–19, but restricting the period to the months between April and September. The upper panel of Fig. 15 shows the mean value of the total number of CG flashes in 10 min over the whole domain (0.9 million km2 wide), as a black line. The trend is clearly negative [trendZ = −0.83 (10 yr)−1], even if the correlation is relatively low (R2 = 0.14) and p value is high (0.17). Also, the 75th, 90th, and 95th percentiles of the mean CG flashes in 10 min yr−1 have decreasing trends, with the highest correlation for the 95%, but with lower values than for the mean value [R2 = 0.11 and trendZ = −0.74 (10 yr)−1]. Again, we can conclude that there is no clear trend, or that there is a statistically weakly significant decreasing trend of CG flashes also in the whole GAR domain, as it was in the FVG plain. It could mean less storms, or the same frequency of storms, but producing fewer lightning flashes per event, on average.
Trends of the EUCLID CG lightning strikes during April–September 2005–19, (top) as mean value and percentiles of the number of strikes every 10 min on the whole GAR domain or (bottom) as number of 10-min cases with at least one strike or above the 90th or 99th percentiles of the total number-of-lightning distributions.
Citation: Journal of Applied Meteorology and Climatology 64, 3; 10.1175/JAMC-D-24-0034.1
To clarify this point, the bottom panel of Fig. 15 shows, as gray bars, the total number of cases having at least one CG lightning in 10 min over the entire domain, i.e., the storm occurrence probability during 2005–19. The fit is very similar to those found in the upper panel [R2 = 0.10 and trendZ = −0.71 (10 yr)−1], i.e., it is the average frequency of storm occurrence over GAR that is decreasing, but with a weak statistical significance.
Considering the tail of the cases with a number of CG flashes above the 90th percentile of the whole distribution, the results are the same. Moreover, if only the 99th percentile (i.e., only the 10 min periods having the largest number of CG flashes) is considered, then a more robust decreasing trend is found: R2 = 0.19 with p value = 0.10 and trendZ = −0.9 (10 yr)−1. In conclusion, also in the GAR domain, it seems that there is a statistically weakly significant decrease in CG flashes during 2005–19, probably due to a small decrease in storm occurrence, that holds also for the very rare (1%) subsample of the strongest 10-min cases.
e. A linear relation between CAPE and lightning density
In the previous section, we analyzed lightning flashes in a much larger domain than FVG. One can be concerned whether there is still a robust relation between the mean number of CG flashes in the GAR domain and the indices derived from the individual sounding made in Udine. We note that NE Italy encompasses the maximum density of CG flashes in the whole GAR (dark areas in Fig. 2); thus, a significant portion of the mean-domain lightning signal is coming just from the NE Italy area. Furthermore, there is a very large linear correlation between the mean-GAR-domain number of CG flashes and the Udine CAPE, considering the mean values per each day of the year (from January to December in this specific section). In fact, if we correlate the mean day-of-year value of CAPE, derived from the Udine 0000 and 1200 UTC soundings, with the mean day-of-year value of GAR total CG flashes during 2005–19 (after taking a ±10 days moving average), it gives a correlation as high as R = 0.98 (R2 = 0.96). It is another confirmation that, from a climatological perspective, lightning occurrence in Europe is mainly a function of CAPE, as stated by Taszarek et al. (2017).
To test if this strong correlation between climatological mean values varies with climate change, we tried to study this relation in two separate periods: first 2005–11 and then 2013–19 (7 years for each period). The black dashed line in Fig. 16 shows the annual trend of GAR LD, while the black solid line is the mean CAPE derived from Udine radiosounding. In the 2005–11 period, from mid-April to November, LD almost always lies above CAPE, whereas the opposite prevails during the 2013–19 period. Hence, in the more recent period, CG, lightnings have a similar day-of-year trend as in the previous one, but the mean CAPE has a much higher one. This confirms that, despite a large increase in instability, there is not a corresponding significant increase in CG flashes, confirming what was previously found.
Annual cycle of the EUCLID CG LD in the GAR domain (dashed black line), together with CAPE derived from the Udine sounding or from ECMWF ERA5 reanalysis (solid black and gray, respectively), and sounding-derived CIN (dot-dashed line) during (top) 2005–11 or (bottom) 2013–19. The A ± 10-day moving average has been applied to all annual cycles.
Citation: Journal of Applied Meteorology and Climatology 64, 3; 10.1175/JAMC-D-24-0034.1
Modification of the linear fit between LD in the GAR domain and CAPE, from the 2005–11 period to the 2013–19 period.
Citation: Journal of Applied Meteorology and Climatology 64, 3; 10.1175/JAMC-D-24-0034.1
5. Summary and conclusions
In this work, about 11 thousand high vertical resolution soundings, made at 0000 and 1200 UTC in Udine (NE Italy) during April–September 1992–2022, were analyzed to derive 70 parameters. Our initial analysis focused on examining the climatological trends of these parameters. One notable finding is that the tropospheric temperature has increased by 1.63°C in 31 years (0.53°C per decade) on average. This temperature increase is more pronounced at midlevels (700 and 500 hPa) compared to 925 hPa. As a consequence of this temperature rise, there has been a 7.7% °C−1 increase in the saturated vapor pressure
Because of the simultaneous effects of warming and an increase in e and q, Θe(p, T, q) increased more than Θes(p, T). For this reason, the potential instability estimated by MaxBuo has increased, as confirmed also by the significant trends of CAPE, DTC, and MULI toward more unstable values. To a lesser extent, also ShowI, lifted index (LI), and K index (KI) evolve toward higher potential instability. Wind shear (up to both 12 and 3 km) also increased, even if the bulk shear between the surface and 850 hPa increased less and the individual magnitudes of wind components have basically no trend (except LLWv). As a consequence of the CAPE increase, the updraft velocity and the estimated maximum hail diameter increased, suggesting an increase in hailstone diameters. Also, the “mixed” indices SWEAT, the “stability and wind shear index for thunderstorms in Switzerland” (SWISS), EHI, and bulk Richardson number (BRI) have trends toward more unstable conditions, even if with very different intensities (much higher for SWEAT than for BRI).
Moreover, the frequency of the potentially unstable cases (i.e., CAPE > 0) is increasing. Indeed, CAPE has a significant trend toward a less skewed distribution, i.e., a longer and heavier right tail and a lower number of null CAPE cases. The—negative—CIN has increased (decreasing in absolute values), and hence, there is less inhibition for convection. Also, the frequency of cases with CIN ≥ −10 J kg−1 has an increasing trend. However, this does not automatically assure that there are more CI events, because the trend of the mean CIN value disappears if one considers only the subset of soundings associated with significant instability (bottom line of Fig. 6). In any case, it does not seem that there is less CI in recent years with respect to the previous years. In fact, for the GAR domain, Manzato et al. (2022a) have found no significant trend of the CI events, objectively identified from CG flash data. Similar results have been found by Augenstein et al. (2023). In fact, studying lightning and thunderstorm days in Europe, they found significant negative trends in central France, in parts of Belgium, and in western and southern Germany, while positive trends were found only for northern Spain. However, no significant trend was found all over the rest of Europe, including the pan-Alpine domain studied by Manzato et al. (2022a).
The cloud-base temperature (parcel temperature at LCL) has a substantial positive trend, while the height of LCL (and of LFC) is slightly decreasing. On the contrary, the cloud top (EL) is strongly increasing, resulting in an even larger increase in the cloud depth (EL-LCL) and of the precipitable water inside the cloud (PWC). In particular, the tail of the highest values of EL-LCL is becoming more populated, demonstrating that the so-called deep convection (e.g., clouds deeper than 9 km) is becoming more frequent nowadays. A lower LCL also means an increase in the in-cloud depth before reaching the freezing level, which means an increase in the cloud “warm layer,” which should favor the warm rain processes, possibly changing some microphysical characteristics. However, a higher EL should also increase the formation of ice inside the cloud, which should favor the hail embryos and the cloud-electrification processes. Also, the height of the 0°C wet-bulb temperature of the cloud (WBZ) had a substantial increase, which favors the melting of hail, particularly of small hailstones that fully melt before reaching the ground. This fact, together with the increase in UpDr and HD, should move the distribution of hailstone diameters toward larger values. The estimated height of the PBL top seems to decrease by about 350 m in these 31 years.
From all these trends, an increase in the frequency and intensity of storms could be expected, with more lightning, hail, and rain. Instead, what is found, with the available data for NE Italy, can be summarized as follows.
-
No increase in rain in FVG, apart from a statistically weak signal of more cases with only 1 mm in 6, 12, and 24 h, which is possibly influenced by an increasing number of sampling stations with time. In fact, in the more coherent dataset of the 26 stations with hourly rain, this feature is not visible. This result agrees with what was found in a recent work by Isotta et al. (2024): They analyzed rain gauge data between 1871 and 2017 in a Pan-Alpine domain, and for NE Italy, a slightly decreasing trend6 of about 1% (10 yr)−1 was found for precipitation during the period March–August (their Figs. 4c,d). In any case, convective rain does not scale with the Clausius–Clapeyron law, at least in FVG, which is one of the regions with the highest rain climatology in central Europe. Note that these results apply only to the April–September rainfalls, while nothing can be said about the rain trends in the other periods of the year, and in particular on the intense “flux” rainfalls occurring in autumn (Manzato 2007), as the “Vaia” event described in Sioni et al. (2023).
-
A tendency toward less hailstorms in the FVG plain, which is not statistically significant, but a statistically significant increasing trend for the median of the hailstone diameters, as suggested also by the increasing trend of HD and WBZ.
-
A statistically significant trend toward lower storm intensities (as estimated by CALCA6h) in the FVG plain. The decrease in CALCA6h is due to a lower frequency of storms; e.g., also the frequency of CALCA6h > 0.6 or >0.8 occurrences shows a—weak—decrease. However, the relative frequency of significant storms with respect to the total number of storms is significantly increasing: In analogy with hail, storms in the FVG plain seem to be less frequent, but the percentage of cases with CALCA6h > 0.7 relative to total storms is higher than it was in the past.
-
No trend of CG lightning flashes in the large GAR domain, or, for some parameters, a statistically weak signal of decreasing CG flashes.
In practice, despite the very significant trends of most environmental parameters, the storm-related observations do not show a corresponding response, at least from the data available to the authors for NE Italy. In conclusion, a statistical relationship, built on the past data of storm-related observations and sounding-derived indices, will likely not be the same in the future, because its climate change invariance is not guaranteed. This was discussed in the example of section 4e, finding that the day-of-year relation between CAPE and lightning density changed between two different 7-yr periods, even if conserving the same high correlation. This indicates that, in NE Italy, convection initiation and storm development are not fully described by the statistical fit between average values. For example, the physical processes leading to lightning formation are much more complex than what CAPE alone can tell us. In fact, the value of CAPE observed by radiosoundings, taken alone, is not very skillful in forecasting storm occurrence/intensity in the following 6 h (Manzato 2003, 2005, 2007, 2012).
Of course, our analysis has some limits, for example, the fact that we analyzed time series of different lengths (some of which are shorter than the usual 30 years used to assess climatological trends) and with different uncertainties, but that is intrinsic to the available data. Moreover, after repeating the analysis of the sounding-derived indices starting from 2002 (i.e., excluding soundings made with the old RS80 sonde), we found a similar ranking of trends as in Table 1 (not shown). However, we strongly believe that, to verify and improve the projections of weather and climate models, it is fundamental to study real observations, using data with the best quality and the highest spatiotemporal resolution. In fact, a comparison with ERA5-derived indices above Udine revealed that PWE and CAPE have similar tendencies, but with a much smaller slope than from the corresponding soundings, while—surprisingly—CIN from ERA5 has an opposite slope (in ERA5 convective inhibition increases with time). This highlights the potential failure of studies based only on model-derived trends that can produce results in contrast with those based on in situ observations. In the case of storm trends, it is not easy to find datasets of observed rain from a dense network, lightning flashes, and hail at ground. For this reason, this study is limited to a small region. However, the authors believe that it would be crucial to perform similar comparisons in other regions worldwide.
Since the mixing ratio at 300 hPa is quite negligible, Θe at that level has been considered to be reliable.
Manzato et al. (2016) manually checked the database of 6-h maximum rainfalls from these 104 stations, from February 2006 to 15 Feb 2015, to improve the data quality. In this work, also the 6-h data from February 2015 to 2022 have been added to the same “high quality database.”
All the cases with more than 50 mm in 1 h were controlled by two experts, comparing with rain in close stations and radar data. A comparison with radar was done also for all cases of ≥20 mm in 1 h when nearby stations showed no rain.
This strong link between updraft and downdraft properties was originally called the “precipitation-generation argument” by Trapp and Woznicki (2017).
During the 2023 European Conference on Severe Storms in Bucharest an ECMWF fellow officially declared that recently ECMWF has changed the way to compute buoyancy, from the relative difference in Θe (between the lifted parcel and the environment) to the relative difference in “virtual Θe.” In SOUND_ANALYS.PY, buoyancy is computed as the relative difference in virtual temperature Tυ.
Trend expressed as a percentage of change in 10 years with respect to the mean rain in the period 1981–2010.
As explained in Bohren and Albrecht (1998), rh was commonly defined as rh = e/esat(T), until the WMO suggested the use of rh = 100 × q/qsat(T, p) = 100 × e × [p − esat(T)]/[(p − e) × esat(T)], where qsat is the saturated mixing ratio. However, the difference is not large.
In this work, we considered only saturation to liquid water and not to ice
Note that if rh0 = e0/esat0 and rh1 = e1/esat1 then
Acknowledgments.
The authors want to thank all the volunteers of the hailpad network; Wolfgang Schulz (ALDIS) for providing the EUCLID data; the OSMER-ARPA FVG personnel working with the data quality check; and Aeronautica Militare for providing the high-resolution Udine soundings. Thanks also to Fulvio Stel (ARPA FVG scientific director), who initially triggered the discussion on the trend of sounding-derived indices.
Data availability statement.
High vertical resolution sounding data are a property of Aeronautica Militare (Italian Air Force) and should be asked at dati.meteo@aeronautica.difesa.it. Rain data are freely available at https://www.osmer.fvg.it/archivio.php?ln=&p=dati. EUCLID CG data are a property of the EUCLID consortium and should be asked for at https://www.euclid.org. The lightning data studied here were given in the frame of the “Multi-Scale Transport and Exchange Processes in the Atmosphere over Mountains Program and Experiment” (TEAMx; http://www.teamx-programme.org). Hail data used in this work are collected by volunteers and are subject to different kinds of errors, and access to individual hailpad data is not available. However, an aggregated version of this dataset is freely available at https://www.meteo.fvg.it/grandine.php.
APPENDIX
How Moisture Trends Scale with Global Warming?
It was shown that, of all the 70 indices studied in this work, all the rh fields are listed in the table of the indices having a low trend (Table 2). In particular, rh at 700 and 500 hPa has among the smallest trends, with a “high p value” (0.20). The “largest” trend for rh (at 850 hPa) has only R2 = 0.30 with p value = 0.02 and trendZ = 0.47 (10 yr)−1. As seen in section 3a, it seems that rh does not change a lot with global warming; i.e., an increase in T corresponds to an increase in absolute moisture q able to maintain an almost constant relative humidity.
It is often said that the atmosphere “moistens” by about 7% °C−1 of warming (Trenberth et al. 2003), due to the Clausius–Clapeyron equation. However, the saturation vapor pressure (esat, which is a function of T only) increases by about 7% °C−1, because of the Clausius–Clapeyron equation, but it is not clear what should happen to rh, which is defined as the ratio between the partial pressure of vapor e and esat.7 Some previous works, like Dai (2006), McCarthy et al. (2009), and Sherwood et al. (2010), already contemplated global warming with zero rh trend, at least in the lower part of the troposphere.
In general, as a response to a warming effect, more evapotranspiration could be expected, which should lead to an increase in e. However, the rate of increase is not clear a priori. In fact, there could be also negative feedbacks, such as an increase in cloud cover, that would decrease the transmission of solar radiation. Moreover, if e increases more than esat, the consequent increase in rh would tend to saturate the air and hence stop the process of further evaporation. It could be a possible mechanism explaining why rh does not increase so much.
Table A1 presents how the linear fits of T,
How the linear fit of different moisture variables, computed at different mandatory levels, changes during April–September of years 1992–2022. Overbar means the average value above the full period. The symbol Δ means the difference between the 1992 value (identified by subscript 0) and the 2022 value, computed on the linear fit. Three variations with respect to T are shown: absolute
The absolute change in temperature is slightly increasing going from 925 up to 500 hPa, while at 300 hPa it is significantly lower. The relative increase in
On the other hand, the relative change in e with respect to the change in T increases much more: from 11.4% °C−1 at 925 hPa up to 12.7% °C−1 at 500 hPa. This suggests that also rh should increase with time. Last, the change in Z(e) compared to Z(T) is slightly decreasing with altitude, going from 1.4 at 925 hPa to 1.2 at 300 hPa, but has a surprising minimum at 500 hPa of 0.9.
Considering the relative change in q with T, it varies from 11.6% to 12.7% °C−1 moving from 925 up to 500 hPa, but has a minimum at 700 hPa of 10.6% °C−1. It could be somewhat related to the fact that at 700 hPa there is the largest ratio of q00/q12 (Fig. 4). It is worth noting that the relative increase in q is about 1.5 times larger than that of
Last, studying the dewpoint temperature, it is found that the ratio of the change in Td with respect to that in T decreases from 1.5 at 925 hPa to 1.1 at 500 hPa, while it jumps to about 2 at 300 hPa, which is probably an unreliable value. In any case, Td increases more than T at any level, but in particular at 925 and 850 hPa. Since the absolute value of Td, expressed in kelvin (K), is a number much larger than those used for the other variables considered in Table A1, we do not show the relative change (normalized by Td0), because it is not comparable with the others (too small). However, it is still possible to compute the ratio for the standardized values, which is not so different from what is already found for e and q, decreasing from 1.4 to 0.8 rising from 925 up to 500 hPa, while at 300 hPa it increases to 1.1.
In conclusion, we can state that from Udine sounding data it seems that, during April–September 1992–2022, T and
REFERENCES
Allen, J. T., 2018: Climate Change and Severe Thunderstorms. Oxford Research Encyclopedia of Climate Science, 67 pp.
Allen, M. R., and W. J. Ingram, 2002: Constraints on future changes in climate and the hydrologic cycle. Nature, 419, 224–232, https://doi.org/10.1038/nature01092.
Auer, I., and Coauthors, 2007: HISTALP—Historical instrumental climatological surface time series of the Greater Alpine Region. Int. J. Climatol., 27, 17–46, https://doi.org/10.1002/joc.1377.
Augenstein, M., S. Mohr, and M. Kunz, 2023: Trends of thunderstorm activity and relation to large-scale atmospheric conditions in western and central Europe. 11th European Conf. on Severe Storms, Bucharest, Romania, European Severe Storms Laboratory, ECSS2023-98, https://doi.org/10.5194/ecss2023-98.
Ban, N., J. Schmidli, and C. Schär, 2015: Heavy precipitation in a changing climate: Does short-term summer precipitation increase faster? Geophys. Res. Lett., 42, 1165–1172, https://doi.org/10.1002/2014GL062588.
Berg, P., C. Moseley, and J. O. Haerter, 2013: Strong increase in convective precipitation in response to higher temperatures. Nat. Geosci., 6, 181–185, https://doi.org/10.1038/ngeo1731.
Betts, A. K., and M. F. Silva Dias, 1979: Unsaturated downdraft thermodynamics in cumulonimbus. J. Atmos. Sci., 36, 1061–1071, https://doi.org/10.1175/1520-0469(1979)036<1061:UDTIC>2.0.CO;2.
Beucler, T., and Coauthors, 2021: Climate-invariant machine learning. arXiv, 2112.08440v5, https://doi.org/10.48550/arXiv.2112.08440.
Bohren, C. F., and B. A. Albrecht, 1998: Atmospheric Thermodynamics. 1st ed. Oxford University Press, 416 pp.
Bolton, D., 1980: The computation of equivalent potential temperature. Mon. Wea. Rev., 108, 1046–1053, https://doi.org/10.1175/1520-0493(1980)108<1046:TCOEPT>2.0.CO;2.
Bony, S., and Coauthors, 2006: How well do we understand and evaluate climate change feedback processes? J. Climate, 19, 3445–3482, https://doi.org/10.1175/JCLI3819.1.
Braca, G., M. Bussettini, B. Lastoria, S. Mariani, and F. Piva, 2021: Il Bilancio Idrologico Gis BAsed a scala Nazionale su Griglia regolare—BIGBANG: Metodologia e stime. Rapporto sulla disponibilità naturale della risorsa idrica (in Italian). ISPRA Rep., 202 pp., https://www.isprambiente.gov.it/it/pubblicazioni/rapporti/il-bilancio-idrologico-gis-based-a-scala-nazionale-su-griglia-regolare-bigbang.
Brooks, H. E., 2013: Severe thunderstorms and climate change. Atmos. Res., 123, 129–138, https://doi.org/10.1016/j.atmosres.2012.04.002.
Brooks, H. E., J. W. Lee, and J. P. Craven, 2003: The spatial distribution of severe thunderstorm and tornado environments from global reanalysis data. Atmos. Res., 67–68, 73–94, https://doi.org/10.1016/S0169-8095(03)00045-0.
Chen, J., A. Dai, Y. Zhang, and K. L. Rasmussen, 2020: Changes in convective available potential energy and convective inhibition under global warming. J. Climate, 33, 2025–2050, https://doi.org/10.1175/JCLI-D-19-0461.1.
Ciccarelli, N., J. von Hardenberg, A. Provenzale, C. Ronchi, A. Vargiu, and R. Pelosini, 2008: Climate variability in north-western Italy during the second half of the 20th century. Global Planet. Change, 63, 185–195, https://doi.org/10.1016/j.gloplacha.2008.03.006.
Dai, A., 2006: Recent climatology, variability, and trends in global surface humidity. J. Climate, 19, 3589–3606, https://doi.org/10.1175/JCLI3816.1.
Davies, J. M., 1993: Hourly helicity, instability, and EHI in forecasting supercell tornadoes. 17th Conf. on Severe Local Storms, Saint Louis, MO, Amer. Meteor. Soc., 107–111.
Davies-Jones, R., 2015: A review of supercell and tornado dynamics. Atmos. Res., 158–159, 274–291, https://doi.org/10.1016/j.atmosres.2014.04.007.
Davies-Jones, R. P., D. Burgess, and M. Foster, 1990: Test of helicity as a tornado forecast parameter. 16th Conf. on Severe Local Storms, Kanaskis Park, Alberta, Canada, Amer. Meteor. Soc., 588–592.
Davolio, S., A. Volonté, A. Manzato, A. Pucillo, A. Cicogna, and M. E. Ferrario, 2016: Mechanisms producing different precipitation patterns over north-eastern Italy: Insights from HyMeX-SOP1 and previous events. Quart. J. Roy. Meteor. Soc., 142, 188–205, https://doi.org/10.1002/qj.2731.
Diffenbaugh, N. S., M. Scherer, and R. J. Trapp, 2013: Robust increases in severe thunderstorm environments in response to greenhouse forcing. Proc. Natl. Acad. Sci. USA, 110, 16 361–16 366, https://doi.org/10.1073/pnas.1307758110.
Durre, I., C. N. Williams Jr., X. Yin, and R. S. Vose, 2009: Radiosonde-based trends in precipitable water over the Northern Hemisphere: An update. J. Geophys. Res., 114, D05112, https://doi.org/10.1029/2008JD010989.
Galway, J. G., 1956: The lifted index as a predictor of latent instability. Bull. Amer. Meteor. Soc., 37, 528–529, https://doi.org/10.1175/1520-0477-37.10.528.
George, J. J., 1960: Weather Forecasting for Aeronautics. Academic Press, 673 pp.
Giaiotti, D., E. Gianesini, and F. Stel, 2001: Heuristic considerations pertaining to hailstone size distributions in the plain of Friuli-Venezia Giulia. Atmos. Res., 57, 269–288, https://doi.org/10.1016/S0169-8095(01)00080-1.
Giaiotti, D., S. Nordio, and F. Stel, 2003: The climatology of hail in the plain of Friuli Venezia Giulia. Atmos. Res., 67–68, 247–259, https://doi.org/10.1016/S0169-8095(03)00084-X.
Giorgi, F., J. W. Hurrell, M. R. Marinucci, and M. Beniston, 1997: Elevation dependency of the surface climate change signal: A model study. J. Climate, 10, 288–296, https://doi.org/10.1175/1520-0442(1997)010<0288:EDOTSC>2.0.CO;2.
Groenemeijer, P. H., and A. van Delden, 2007: Sounding-derived parameters associated with large hail and tornadoes in the Netherlands. Atmos. Res., 83, 473–487, https://doi.org/10.1016/j.atmosres.2005.08.006.
Hersbach, H., and Coauthors, 2020: The ERA5 global reanalysis. Quart. J. Roy. Meteor. Soc., 146, 1999–2049, https://doi.org/10.1002/qj.3803.
Hirt, M., G. C. Craig, S. A. K. Schäfer, J. Savre, and R. Heinze, 2020: Cold-pool-driven convective initiation: using causal graph analysis to determine what convection-permitting models are missing. Quart. J. Roy. Meteor. Soc., 146, 2205–2227, https://doi.org/10.1002/qj.3788.
Huntrieser, H., H. H. Schiesser, W. Schmid, and A. Waldvogel, 1997: Comparison of traditional and newly developed thunderstorm indices for Switzerland. Wea. Forecasting, 12, 108–125, https://doi.org/10.1175/1520-0434(1997)012<0108:COTAND>2.0.CO;2.
IPCC, 2001: Climate Change 2001: The Scientific Basis. J. T. Houghton et al., Eds., Cambridge University Press, 881 pp.
Isotta, F. A., and Coauthors, 2014: The climate of daily precipitation in the Alps: Development and analysis of a high-resolution grid dataset from pan-Alpine rain-gauge data. Int. J. Climatol., 34, 1657–1675, https://doi.org/10.1002/joc.3794.
Isotta, F. A., Chimani, B., Hiebl, J., and C. Frei. 2024: Long-term Alpine precipitation reconstruction (LAPrec): A gridded monthly data set dating back to 1871. J. Geophys. Res. Atmos., 129, e2023JD039637, https://doi.org/10.1029/2023JD039637.
Kahraman, A., E. J. Kendon, and H. J. Fowler, 2024: Climatology of severe hail potential in Europe based on a convection-permitting simulation. Climate Dyn., 62, 6625–6642, https://doi.org/10.1007/s00382-024-07227-w.
Kopp, J., A. Manzato, A. Hering, U. Germann, and O. Martius, 2023: How observations from automatic hail sensors in Switzerland shed light on local hailfall duration and compare with hailpad measurements. Atmos. Meas. Tech., 16, 3487–3503, https://doi.org/10.5194/amt-16-3487-2023.
Kunz, M., J. Wandel, E. Fluck, S. Baumstark, S. Mohr, and S. Schemm, 2020: Ambient conditions prevailing during hail events in central Europe. Nat. Hazards Earth Syst. Sci., 20, 1867–1887, https://doi.org/10.5194/nhess-20-1867-2020.
Leutwyler, D., O. Fuhrer, and X. Lapillonne, D. Lüthi, and C. Schär, 2016: Towards European-scale convection-resolving climate simulations with GPUs: A study with COSMO 4.19. Geosci. Model Dev., 9, 3393–3412, https://doi.org/10.5194/gmd-9-3393-2016.
Lucarini, V., A. Speranza, and R. Vitolo, 2007: Parametric smoothness and self-scaling of the statistical properties of a minimal climate model: What beyond the mean field theories? Physica D, 234, 105–123, https://doi.org/10.1016/j.physd.2007.07.006.
Mallinson, H., S. Lasher-Trapp, J. Trapp, M. Woods, and S. Orendorf, 2023: Hailfall in a possible future climate using a pseudo–global warming approach: Hail characteristics and mesoscale influences. J. Climate, 37, 527–549, https://doi.org/10.1175/JCLI-D-23-0181.1.
Manzato, A., 2003: A climatology of instability indices derived from Friuli Venezia Giulia soundings, using three different methods. Atmos. Res., 67–68, 417–454, https://doi.org/10.1016/S0169-8095(03)00058-9.
Manzato, A., 2005: The use of sounding-derived indices for a neural network short-term thunderstorm forecast. Wea. Forecasting, 20, 896–917, https://doi.org/10.1175/WAF898.1.
Manzato, A., 2007: The 6 h climatology of thunderstorms and rainfalls in the Friuli Venezia Giulia plain. Atmos. Res., 83, 336–348, https://doi.org/10.1016/j.atmosres.2005.08.013.
Manzato, A., 2008: A verification of numerical model forecasts for sounding-derived indices above Udine, northeast Italy. Wea. Forecasting, 23, 477–495, https://doi.org/10.1175/2007WAF2007018.1.
Manzato, A., 2012: Hail in northeast Italy: Climatology and bivariate analysis with the sounding-derived indices. J. Appl. Meteor. Climatol., 51, 449–467, https://doi.org/10.1175/JAMC-D-10-05012.1.
Manzato, A., 2013: Hail in northeast Italy: A neural network ensemble forecast using sounding-derived indices. Wea. Forecasting, 28, 3–28, https://doi.org/10.1175/WAF-D-12-00034.1.
Manzato, A., and G. M. Morgan Jr., 2003: Evaluating the sounding instability with the Lifted Parcel Theory. Atmos. Res., 67–68, 455–473, https://doi.org/10.1016/S0169-8095(03)00059-0.
Manzato, A., Cicogna, A., and A. Pucillo, 2016: 6-hour maximum rain in Friuli Venezia Giulia: Climatology and ECMWF-based forecasts. Atmos. Res., 169B, 465–484, https://doi.org/10.1016/j.atmosres.2015.07.013.
Manzato, A., S. Serafin, M. M. Miglietta, D. Kirshbaum, and W. Schulz, 2022a: A pan-Alpine climatology of lightning and convective initiation. Mon. Wea. Rev., 150, 2213–2230, https://doi.org/10.1175/MWR-D-21-0149.1.
Manzato, A., A. Cicogna, M. Centore, P. Battistutta, and M. Trevisan, 2022b: Hailstone characteristics in Northeast Italy from 29 years of hailpad data. J. Appl. Meteor. Climatol., 61, 1779–1795, https://doi.org/10.1175/JAMC-D-21-0251.1.
Markowski, P. M., and Y. P. Richardson, 2009: Tornadogenesis: Our current understanding, forecasting considerations, and questions to guide future research. Atmos. Res., 93, 3–10, https://doi.org/10.1016/j.atmosres.2008.09.015.
Martinkova, M., and J. Kysely, 2020: Overview of observed Clausius-Clapeyron scaling of extreme precipitation in midlatitudes. Atmosphere, 11, 786, https://doi.org/10.3390/atmos11080786.
Mattar, C., J. A. Sobrino, Y. Julien, and L. Morales, 2011: Trends in column integrated water vapour over Europe from 1973 to 2003. Int. J. Climatol., 31, 1749–1757, https://doi.org/10.1002/joc.2186.
McCarthy, M. P., P. W. Thorne, and H. A. Titchner, 2009: An analysis of tropospheric humidity trends from radiosondes. J. Climate, 22, 5820–5838, https://doi.org/10.1175/2009JCLI2879.1.
Micheletti, S., and M. Salvador, 2014: Ricognizione reti meteoclimatiche regionali (In Italian). OSMER-ARPA FVG Tech. Rep. v3.2, 270 pp.
Miglietta, M. M., A. Manzato, and R. Rotunno, 2016: Characteristics and predictability of a supercell during HyMeX SOP1. Quart. J. Roy. Meteor. Soc., 142, 2839–2853, https://doi.org/10.1002/qj.2872.
Miller, R. C., 1972: Notes on the analysis and severe-storm forecasting procedures of the Air Force Global Weather Central. Air Weather Service Tech. Rep. 200 (Rev.), Air Weather Service, Scott Air Force Base, 190 pp.
Miloshevich, L. M., H. Vömel, D. N. Whiteman, B. M. Lesht, F. J. Schmidlin, and F. Russo, 2006: Absolute accuracy of water vapor measurements from six operational radiosonde types launched during AWEX-G and implications for AIRS validation. J. Geophys. Res., 111, D09S10, https://doi.org/10.1029/2005JD006083.
Miloshevich, L. M., H. Vömel, D. N. Whiteman, and T. Leblanc, 2009: Accuracy assessment and correction of Vaisala RS92 radiosonde water vapor measurements. J. Geophys. Res., 114, D11305, https://doi.org/10.1029/2008JD011565.
Mohr, S., and M. Kunz, 2013: Recent trends and variabilities of convective parameters relevant for hail events in Germany and Europe. Atmos. Res., 123, 211–228, https://doi.org/10.1016/j.atmosres.2012.05.016.
Mohr, S., M. Kunz, and K. Keuler, 2015: Development and application of a logistic model to estimate the past and future hail potential in Germany. J. Geophys. Res. Atmos., 120, 3939–3956, https://doi.org/10.1002/2014JD022959.
Molnar, P., S. Fatichi, L. Gaál, J. Szolgay, and P. Burlando, 2015: Storm type effects on super Clausius–Clapeyron scaling of intense rainstorm properties with air temperature. Hydrol. Earth Syst. Sci., 19, 1753–1766, https://doi.org/10.5194/hess-19-1753-2015.
Moradi, I., B. Soden, R. Ferraro, P. Arkin, and H. Vömel, 2013: Assessing the quality of humidity measurements from global operational radiosonde sensors. J. Geophys. Res. Atmos., 118, 8040–8053, https://doi.org/10.1002/jgrd.50589.
Morgan, G. M., 1992: Some results of aircraft investigation of internal properties of thunderstorm cloud systems in northeastern Italy with an interpretation for hail prevention. Atmos. Res., 28, 259–269, https://doi.org/10.1016/0169-8095(92)90012-Y.
Morgan, G. M., Jr., and N. G. Towery, 1975: Small-scale variability of hail and its significance for hail prevention experiments. J. Appl. Meteor., 14, 763–770, https://doi.org/10.1175/1520-0450(1975)014<0763:SSVOHA>2.0.CO;2.
Morgan, G. M., Jr., and J. Tuttle, 1984: Some experimental techniques for the study of the evolution of atmospheric thermodynamic instability. Proc. Second Int. Conf. on Hailstorms and Hail Suppression, Sofia, Bulgaria, Bulgarian Hydrometeorological Service, 192–196.
Neelin, J. D., C. Martinez-Villalobos, S. N. Stechmann, F. Ahmed, G. Chen, J. M. Norris, Y.-H. Kuo, and G. Lenderink, 2022: Precipitation extremes and water vapor. Curr. Climate Change Rep., 8, 17–33, https://doi.org/10.1007/s40641-021-00177-z.
O’Gorman, P. A., and T. Schneider, 2009: Scaling of precipitation extremes over a wide range of climates simulated with an idealized GCM. J. Climate, 22, 5676–5685, https://doi.org/10.1175/2009JCLI2701.1.
Pall, P., M. R. Allen, and D. A. Stone, 2007: Testing the Clausius–Clapeyron constraint on changes in extreme precipitation under CO2 warming. Climate Dyn., 28, 351–363, https://doi.org/10.1007/s00382-006-0180-2.
Pavan, V., and Coauthors, 2019: High resolution climate precipitation analysis for north-central Italy, 1961–2015. Climate Dyn., 52, 3435–3453, https://doi.org/10.1007/s00382-018-4337-6.
Purdom, J. F. W., 1976: Some uses of high-resolution GOES imagery in the mesoscale forecasting of convection and its behavior. Mon. Wea. Rev., 104, 1474–1483, https://doi.org/10.1175/1520-0493(1976)104<1474:SUOHRG>2.0.CO;2.
Rasmussen, E. N., and D. O. Blanchard, 1998: A baseline climatology of sounding-derived supercell and tornado forecast parameters. Wea. Forecasting, 13, 1148–1164, https://doi.org/10.1175/1520-0434(1998)013<1148:ABCOSD>2.0.CO;2.
Ross, R. J., and W. P. Elliott, 2001: Radiosonde-based Northern Hemisphere tropospheric water vapor trends. J. Climate, 14, 1602–1612, https://doi.org/10.1175/1520-0442(2001)014<1602:RBNHTW>2.0.CO;2.
San Segundo, H. S., J. A. López, N. Pineda, P. Altube, and J. Montanyà, 2020: Sensitivity analysis of lightning stroke-to-flash grouping criteria. Atmos. Res., 242, 105023, https://doi.org/10.1016/j.atmosres.2020.105023.
Sherwood, S. C., R. Roca, T. M. Weckwerth, and N. G. Andronova, 2010: Tropospheric water vapor, convection, and climate. Rev. Geophys., 48, RG2001, https://doi.org/10.1029/2009RG000301.
Showalter, A. K., 1953: A stability index for thunderstorm forecasting. Bull. Amer. Meteor. Soc., 34, 250–252.
Simon, T., G. J. Mayr, D. Morgenstern, N. Umlauf, and A. Zeileis, 2023: Amplification of annual and diurnal cycles of alpine lightning. Climate Dyn., 61, 4125–4137, https://doi.org/10.1007/s00382-023-06786-8.
Sioni, F., S. Davolio, F. Grazzini, and L. Giovannini, 2023: Revisiting the atmospheric dynamics of the two century floods over north-eastern Italy. Atmos. Res., 286, 106662, https://doi.org/10.1016/j.atmosres.2023.106662.
Skliris, N., J. D. Zika, G. Nurser, S. A. Josey, and R. Marsh, 2016: Global water cycle amplifying at less than the Clausius-Clapeyron rate. Sci. Rep., 6, 38752, https://doi.org/10.1038/srep38752.
Stipanuk, G. S., 1973: Algorithms for generating a Skew-T, log p diagram and computing selected meteorological quantities. Atmospheric Sciences Laboratory, U.S. Army Electronics Command, White Sands Missile Range Tech. Rep. ECOM-5515, 41 pp., https://apps.dtic.mil/sti/tr/pdf/AD0769739.pdf.
Taszarek, M., H. E. Brooks, and B. Czernecki, 2017: Sounding-derived parameters associated with convective hazards in Europe. Mon. Wea. Rev., 145, 1511–1528, https://doi.org/10.1175/MWR-D-16-0384.1.
Taszarek, M., and Coauthors, 2019: A climatology of thunderstorms across Europe from a synthesis of multiple data sources. J. Climate, 32, 1813–1837, https://doi.org/10.1175/JCLI-D-18-0372.1.
Taszarek, M., J. T. Allen, T. Púčik, K. A. Hoogewind, and H. E. Brooks, 2020: Severe convective storms across Europe and the United States. Part II: ERA5 environments associated with lightning, large hail, severe wind, and tornadoes. J. Climate, 33, 10 263–10 286, https://doi.org/10.1175/JCLI-D-20-0346.1.
Taszarek, M., J. T. Allen, M. Marchio, and H. E. Brook, 2021: Global climatology and trends in convective environments from ERA5 and rawinsonde data. npj Climate Atmos. Sci., 4, 35, https://doi.org/10.1038/s41612-021-00190-x.
Tompkins, A. M., 2001: Organization of tropical convection in low vertical wind shears: The role of water vapor. J. Atmos. Sci., 58, 529–545, https://doi.org/10.1175/1520-0469(2001)058<0529:OOTCIL>2.0.CO;2.
Torri, G., and Z. Kuang, 2016: A Lagrangian study of precipitation-driven downdrafts. J. Atmos. Sci., 73, 839–854, https://doi.org/10.1175/JAS-D-15-0222.1.
Torri, G., Z. Kuang, and Y. Tian, 2015: Mechanisms for convection triggering by cold pools. Geophys. Res. Lett., 42, 1943–1950, https://doi.org/10.1002/2015GL063227.
Trapp, R. J., and K. A. Hoogewind, 2016: The realization of extreme tornadic storm events under future anthropogenic climate change. J. Climate, 29, 5251–5265, https://doi.org/10.1175/JCLI-D-15-0623.1.
Trapp, R. J., and J. M. Woznicki, 2017: Convectively induced stabilizations and subsequent recovery with supercell thunderstorms during the Mesoscale Predictability Experiment (MPEX). Mon. Wea. Rev., 145, 1739–1754, https://doi.org/10.1175/MWR-D-16-0266.1.
Trapp, R. J., K. A. Hoogewind, and S. Lasher-Trapp, 2019: Future changes in hail occurrence in the United States determined through convection-permitting dynamical downscaling. J. Climate, 32, 5493–5509, https://doi.org/10.1175/JCLI-D-18-0740.1.
Trenberth, K. E., A. Dai, R. M. Rasmussen, and D. B. Parsons, 2003: The changing character of precipitation. Bull. Amer. Meteor. Soc., 84, 1205–1218, https://doi.org/10.1175/BAMS-84-9-1205.
Tudoroiu, M., E. Eccel, B. Gioli, D. Gianelle, H. Schume, L. Genesio, and F. Miglietta, 2016: Negative elevation-dependent warming trend in the Eastern Alps. Environ. Res. Lett., 11, 044021, https://doi.org/10.1088/1748-9326/11/4/044021.
Vergara-Temprado, J., N. Ban, and C. Schär, 2021: Extreme sub-hourly precipitation intensities scale close to the Clausius-Clapeyron rate over Europe. Geophys. Res. Lett., 48, e2020GL089506, https://doi.org/10.1029/2020GL089506.
Vömel, H., and Coauthors, 2007: Radiation dry bias of the Vaisala RS92 humidity sensor. J. Atmos. Oceanic Technol., 24, 953–963, https://doi.org/10.1175/JTECH2019.1.
Weisman, M. L., and J. B. Klemp, 1982: The dependence of numerically simulated convective storms on vertical wind shear and buoyancy. Mon. Wea. Rev., 110, 504–520, https://doi.org/10.1175/1520-0493(1982)110<0504:TDONSC>2.0.CO;2.
Westra, S., and Coauthors, 2014: Future changes to the intensity and frequency of short-duration extreme rainfall. Rev. Geophys., 52, 522–555, https://doi.org/10.1002/2014RG000464.
Windmiller, J. M., J. Bao, S. C. Sherwood, T. D. Schanzer, and D. Fuchs, 2023: Predicting convective downdrafts from updrafts and environmental conditions in a global storm resolving simulation. J. Adv. Model. Earth Syst., 15, e2022MS003048, https://doi.org/10.1029/2022MS003048.
Yano, J.-I., and A. Manzato, 2022: “Moisture control” of the precipitation: A probabilistic perspective. Precipitation Science, S. Michaelides, Ed., Elsevier, 615–634, https://doi.org/10.1016/B978-0-12-822973-6.00018-4.
