Reforecasting the November 1994 flooding of Piedmont with a convection-permitting model
aa r X i v : . [ phy s i c s . a o - ph ] A ug Bulletin of Atmospheric Science and Technology manuscript No. (will be inserted by the editor)
Reforecasting the November 1994 flooding of Piedmont witha convection-permitting model
Valerio Capecchi
Received: date / Accepted: date
Abstract
The Piedmont region in Italy was affected by a heavy rainfall event inNovember 1994. On the 4th convective cells involved the coastal mountains of theregion. On the 5th and early 6th, there were abundant precipitations, related to oro-graphic lift and low-level convergences, in the Alpine area. This study aims to evalu-ate whether a convection-permitting model provides more valuable information withrespect to past numerical experiments. Results for the 4th of November show that thecloud-resolving model successfully reconstructs the structure of precipitation sys-tems on the downstream side of the coastal mountains. As regards the precipitationsof the 5th of November, no added value is found. However, we provide evidence ofthe anomalously intense transport of moist air from the tropical and subtropical At-lantic and postulate how such transport is responsible for reducing the stability of theflow impinging on the Alps.
Keywords
Convection-permitting model · Severe weather · Past events · Reforecast
Every year, weather related disasters cause huge damage, significant economic lossesand often casualties. According to the “Atlas of Mortality and Economic Losses fromWeather, Climate and Water Extremes 1970-2012” edited by the World Meteoro-logical Organization, floods and storms are the costliest events in Europe. Amongsuch dramatic cases, the severe weather event of November 1994 in the Piedmontregion (north-western Italy) is definitively one to be considered as remarkable. As aconsequence of the heavy rainfall, several rivers in the southern part of the regionflooded causing huge economic losses (estimated at about 14 billion US dollars) and
Valerio CapecchiLaMMA - Laboratorio di Meteorologia e Modellistica Ambientale per lo sviluppo sostenibileVia Madonna del Piano 10, Sesto Fiorentino, Firenze, ItaliaE-mail: [email protected] Capecchi V
70 people died. Past numerical simulations were able to provide quite accurate re-constructions of the event. Nevertheless, the improvements achieved over the last 25years in weather modelling pose the question of whether we can obtain new insightsinto the Piedmont flooding by using convective-scale numerical models.Petroliagis et al. (1996) were the first to study the November 1994 Piedmont case(hereinafter P94). They used the ECMWF global model operational at that time, bothwith the single deterministic forecast and with an ensemble approach. The spectralresolution of their deterministic run (ensemble members) was T213 (T63), whichroughly corresponds to about 60 (210) km grid spacing at the Equator. Despite therough resolution of the model and the fact that the number of members (32 + thecontrol member) of the ensemble was fewer than the one currently used, the authorsreached some conclusions about the predictability of the event. Firstly, the day-5 today-1 T213 forecasts were skilful, with the precipitation patterns well positioned ac-cording to observations. Secondly, the day-1 T213 forecast provided a rainfall maxi-mum of 135 mm , which, although strongly underestimated with respect to observeddata, gave a clear warning of potential severe weather. Finally, the evolution of rainfallpredictions in the day-8 to day-3 T63 simulations exhibited consistency, suggestinga significant possibility of an extreme event in the grid point closest to the area ofinterest. The authors concluded that the probabilistic T63 predictions supported theresults of the deterministic T213 forecast and reinforced the degree of confidence thatcould be associated with it.Analysing the data registered by automatic weather stations, on the 4th of Novem-ber intense rainfall rates affected southern Piedmont, in an area between the MaritimeAlps and Ligurian Apennines (Lionetti 1996; Jansa et al. 2000; Cassardo et al. 2002).Buzzi et al. (1998) noted how large errors in their numerical experiments were asso-ciated to precipitation forecasts in southern Piedmont, where convection was the maincontributor to rainfall. They concluded that these errors were likely due to the lackof an explicit computation of the trajectories of convective cells. A misplacement ofprecipitation maxima in this area was also found in the work by Ferretti et al. (2000).In another paper, Buzzi and Foschini (2000) implemented a high-resolution modelsimulation of the P94 case with a mesh size of 4 km, but without the explicit compu-tation of convection processes. Also in this case, the authors demonstrated how theprecipitation maxima over southern Piedmont were underestimated and misplaced byabout 25-30 km to the south of the actual position. As a consequence, the drainagebasins of the Tanaro, Bormida and Belbo Rivers in southern Piedmont would not beaffected by rainfall runoff with likely errors in hydrological predictions.Romero et al. (1998) used the hydrostatic mesoscale model described in Nickerson et al.(1986) to analyse the P94 case. Numerical simulations agreed well with observedprecipitation patterns, however maximum rainfall predictions were strongly underes-timated. By looking at the convective part of the simulated precipitation, the authorsargued that the rainfall over Piedmont on the 5th of November had a non-convectivenature.This assessment was confirmed by Doswell III et al. (1998) who analysed satelliteimages. They concluded that much of the rainfall over the foothills of the Alps in thePiedmont area was associated with upslope flow of weakly stable, moist air. Further-more, using the ECMWF analysis data, they calculated the Froude number, defined eforecasting the 1994 Piedmont flood 3 by the relationship Fr = UNh , (1)where U is the horizontal wind speed averaged over the 1000-, 925-, and 850-hPaisobaric levels, N is the stability Brunt-V¨ais¨al¨a frequency calculated from the poten-tial temperature difference between 1000- and 850-hPa levels and h is the height ofthe orography (which is about 3000 m in the present case). When the Froude num-ber is greater than 1, flow over the orographic obstacle is favoured, thanks to stronglow-level winds U or to reduced values of the static stability N (or a combinationof the two). On the contrary, the ascent is reduced when Fr ≪
1, favouring flowaround the obstacle (depending on the geometry of the mountains). Doswell III et al.(1998) found that the Froude number was about 0.90 on the 5th of November. Theyconcluded that in this regime, the uplift flow could be attained only in saturated con-ditions.In their idealized numerical study with an L-shape orography (rotated 90 degreesclock-wise), Rotunno and Ferretti (2001) demonstrated how westward deflection ofthe southerly sub-saturated flow blocked by the eastern Alps enhanced low-levelconvergence with the saturated flow directed towards the western Alps. As a con-sequence, this latter flow was forced to ascend further upstream the mountainousbarrier, due to the difference of equivalent potential temperature between the two air-flows (see Figure 16 in their paper).Similar considerations are found in Buzzi et al. (1998). To demonstrate how a rel-atively stable and moist flux can change the Froude number, they conducted someexperiments to assess the role of latent heat exchanges due to condensation and evap-oration. A numerical investigation was conducted by suppressing latent heat releasedue to condensation. Results showed that the low-level flow over the MediterraneanSea was diverted to the western side of the Alps with respect to the control simulation.As a consequence, the precipitation maxima were found in southern France (see Fig-ures 12 and 13 in their paper). To explain such behaviour, the authors underlined thatsuppressing the release of latent heat due to condensation, resulted in higher MSLPvalues in the Alpine region and thus a reduced uplift flow over the mountains. Similarfindings and discussions are reported in Ferretti et al. (2000).Moreover Buzzi et al. (1998) and other authors (Romero et al. 1998; Ferretti et al.2000; Jansa et al. 2000; Cassardo et al. 2002) performed numerical experiments bysuppressing (or reducing) the orography of the model grid. All of them found thatthe precipitation maxima were strongly reduced. Summarising the results in the lit-erature showed that on the 5th of November, the precipitation in the Alps (northernPiedmont) was the result of the combined action of orographic uplift and latent heatexchanges and that convection played a marginal role.Table 1 lists past numerical experiments dealing with the P94 case. All the modelsdeployed some kind of parametrization scheme for convection processes. Mesh sizesranged between 60 km for the global model used in Petroliagis et al. (1996) up to 4km, which is the resolution of the experiments by Buzzi and Foschini (2000).An analysis of the P94 case can be conducted in light of an atmospheric river(AR) landfall. Since the seminal work by Zhu and Newell (1998), AR theory andfield experiments has received much attention in relatively recent years (see for in-
Capecchi V
Table 1
Past numerical investigations on the November 1994 Piedmont caseReference Model Resolution Vertical Forecast(grid spacing) levels lengthPetroliagis et al. (1996) IFS (EPS) T213 (T63) 31 (19) 120 oursRomero et al. (1998) SALSA 20 km 30 30 hoursBuzzi et al. (1998) BOLAM 30 and 10 km 36 36 hoursBuzzi and Foschini (2000) BOLAM up to 4 km 40 30 hoursJansa et al. (2000) HIRLAM 40 km 31 48 hoursFerretti et al. (2000) MM5 30 and 10 km 23 48 hoursCassardo et al. (2002) RAMS 15 km 35 84 hours stance Ralph et al. 2004, Lavers and Villarini 2013 and Krichak et al. 2016). Roughlyspeaking, ARs are defined as narrow tongues of moist air in the lower troposphereresponsible for the transport of tropical water vapour into the extratropics. AlthoughARs are mainly studied for their impacts on the western coasts of the United States,some papers demonstrated that heavy precipitation and flood events in Europe are of-ten linked to ARs landfall (Stohl et al. 2008, Lavers and Villarini 2013, Krichak et al.2016). Furthermore, as stated in Krichak et al. (2016), the regions mostly prone to theimpacts of ARs landfall are mountainous, providing the necessary uplift for signifi-cant rainfall.Over the last few decades, convective-scale numerical weather prediction modellingimproved tremendously (Sun et al. 2014) and is now facing new challenges (Yano et al.2018). Nevertheless the predictability of small-scale, high-impact weather remainslimited due to (i) approximations in the reconstruction of fine-scale processes (Leutbecher et al.2017), and (ii) the chaotic nature of the weather system causing small errors in theinitial conditions to grow rapidly (Palmer 2001). Nowadays, technological capabil-ity allows operational convection-permitting models to be run with horizontal meshsizes less than 2 km as well as ensemble systems with horizontal mesh size about 3km. To mention a few examples in national weather services in Europe, the mesh sizeof the operational model of M´et´eo-France (AROME) is about 1.3 km; the UK MetOffice runs a 1.5 km grid length model four times per day and has plans to deploy asub-kilometre scale version in the near future. The newest version of the ICON modelat the German DWD weather service is a seamless model that can be run with a meshsize of less than 1 km. As regards probabilistic prediction systems, examples includeCOSMO-DE at 2.8 km horizontal resolution (Peralta et al. 2012), MOGREPS-UK at2.2 km horizontal resolution (Hagelin et al. 2017) and AROME-EPS at 2.5 horizontalresolution (Raynaud and Bouttier 2016).The goal of this paper is to perform a reforecast of the P94 case. We presenta numerical reconstruction by applying cutting-edge regional convection-permittingNWP modelling, fed by recently updated global analyses. The purpose is twofold: asregards the rainfall observed on the 4th of November, the challenge is to improve theprediction of the convective precipitations in southern Piedmont, where the majorityof damage and casualties occurred. As regards the rainfall observed on the 5th ofNovember, we provide evidence of an AR landfall during the days of the P94 case. Inaddition, we investigate if a better quantitative precipitation forecast can be accom- eforecasting the 1994 Piedmont flood 5 plished by deploying a high-resolution model, which uses an accurate description oftopography.We conclude this introduction by stressing the fact that evaluating whether andhow current state-of-the-art numerical models can simulate past high-impact eventsis essential to understand the information content of current operational forecastingsystems in order to predict future extreme events.
Although detailed synoptic descriptions of the P94 case can be found in the pa-pers listed in Table 1, we include here a short summary to make this paper self-contained. October 1994 was characterised by a remarkable weather variability overnorth-western Italy. Accumulated rainfall recorded by automatic weather stationswas above average in several districts of Piedmont. Between the 2nd and 4th ofNovember, there was an upper level cyclonic circulation over the northern Atlanticarea (Petroliagis et al. 1996). ECMWF analysis data valid at 12 UTC of the 4th ofNovember showed a trough with an axis extending from the British Isles to theIberian peninsula (see Figure 3b in Buzzi et al. 1998). This configuration activatedthe advection of warm and moist air over the western/central Mediterranean Sea to-wards southern France and northern Italy. As recorded by rain-gauges data (Lionetti1996), this time practically coincided with the beginning of the heavy precipitationsin southern Piedmont. For instance, see in Figure 1a the accumulated precipitationrecorded at the Ponzone rain-gauge (whose location is indicated in Figure 2). Oneday later, the movement of the trough eastwards was slow, due to the ridge presentover the central/eastern Mediterranean Sea (see Figures 4b in Buzzi et al. 1998 and1a in Petroliagis et al. 1996). As a consequence, a double front-like structure formedwest of Italy as confirmed by numerical simulations and Meteosat satellite images(Buzzi et al. 1998; Doswell III et al. 1998). This configuration is not uncommon inthe western Mediterranean and is often associated, as in the P94 case, with intenseprecipitations in the Alpine region; see the plots in Figure 4 of Lionetti (1996) andthe accumulated precipitation recorded at Oropa rain-gauge in Figure 1b. We notehow rainfall rates at Oropa reached 27 mm hour − with an average value on the 5thof November about 10 mm hour − . Instead, data recorded at Ponzone (Figure 1a) re-ported a maximum value of 38 mm hour − with about 135 mm in the last four hoursof the 4th of November. Thus the profiles of precipitation rates of Ponzone rain-gauge(southern Piedmont) exhibited much higher time variability, indicating the presenceof convection. This assessment is also found in Buzzi et al. (1998).We conclude this brief overview of the P94 case by stressing the fact that, as demon-strated by the numerical experiments conducted by Romero et al. (1998) and Cassardo et al.(2002), the contribution to the precipitations of the Mediterranean sea surface evapo-ration was likely not important. Capecchi V
Fig. 1
Accumulated precipitations recorded in the period from 00 UTC 4th November 1994 to 12 UTC6th November 1994 by the rain-gauges at Ponzone and Oropa, whose geographical positions are shown inFigure 2 α scale ( ≃ km ) down to the micro- γ scale ( ≃ m ), typical of the Large EddySimulation (LES) models. For a general overview of the Meso-NH model and itsapplications see Lafore et al. (1997) and Lac et al. (2018), while the scientific doc-umentation is available on the model’s website. For this study, we used the version5.4.1 (released in July 2018). The geographical extent of the simulations is shownin Figure 2; no grid-nesting was implemented. As regards microphysics, we set theone-momentum ICE3 scheme (Caniaux et al. 1994), that takes into account five wa-ter species (cloud droplets, raindrops, pristine ice crystals, snow or aggregates, andgraupel). The convection, both deep and shallow, was explicitly resolved. The Runge-Kutta centred 4th order scheme was chosen for momentum advection. This scheme isrecommended when using, as in the present paper, the CENT4TH (4th order CENtred eforecasting the 1994 Piedmont flood 7 Table 2
Key parameters of the Meso-NH model settings
Variable Value
Rows × Columns 300 × Fig. 2
Orography of the domain of the Meso-NH simulation. Horizontal grid spacing ∆ x is 2.5 km. Thelocations of nine rain-gauges are indicated by black dots: three in southern Piedmont and six in the northof the region, at the foothills of the Alps. The accumulated rainfall recorded at Oropa (“Oro”) and Ponzone(“Pon”) are shown in Figure 1. The political borders of the Piedmont region are indicated by the black line on space and time) advection scheme. The CENT4TH was chosen because of its nu-merical stability, although it is more time-consuming than other options (Lunet et al.2017). Some specific parameters of the model for this study are summarised in Table2. To drive the Meso-NH simulations, initial and boundary conditions were pro-duced using a recent version of the ECMWF-IFS model (cycle 41r2, the same usedto produce the ERA5 data). The spectral resolution of the model is TL1279, whichcorresponds to about 16 km grid spacing. Boundary conditions are provided every 3hours. Capecchi V
IV T = s(cid:18) g Z p top p qu d p (cid:19) + (cid:18) g Z p top p qv d p (cid:19) where q ( p ) is the specific humidity in KgKg − , u ( p ) and v ( p ) are the zonal andmeridional components respectively of the horizontal wind vector in ms − , g is theacceleration due to gravity, p and p top are the 1000- and 300-hPa isobaric levelsrespectively. The algorithms based on this approach (Rutz et al. 2014) declare gridpoints as interested by an AR if they satisfy the condition that IVT exceeds a pre-defined threshold, which is normally equal to 250 kg m − s − . Ralph et al. (2019)proposed a scale to characterize the strength and impacts of ARs. It is based on theanalysis of both the maximum value of IVT at a given location during the AR event(i.e. IV T ≥ kg m − s − ) and the AR duration. The IVT maps presented belowwere calculated using the ERA5 data (Hersbach et al. 2019). In Figure 3 we show the 24-hour accumulated precipitations predicted by the Meso-NH model for the 4th of November. Observed amounts (coloured circles) overlap theforecasts data. Starting time of the simulation is 00 UTC 4 November 1994. Fore-casts initialised one and two days in advance do not provide substantial differences tothe one shown here. From a visual inspection of Figure 3 the pattern of precipitationforecast is in good agreement with observed values. However, the simulation tendsto slightly overestimate rainfall along the coastal mountains (that is in southern Pied-mont). As regards the comparison with observations collected at Ponzone rain-gauge,the maximum rainfall is misplaced by about 20 km. In Figure 4 we show the verticalcross section at 13 UTC of the 4th of November of a transect (indicated in panel 4a)close to the Ponzone rain-gauge. The panels 4b, 4c and 4d show the mixing ratio ofcloud droplets, rain and graupel respectively and the wind vectors at model levels.Convection is visible on the lee side of the coastal mountains associated to moderateto strong vertical velocities. The model reconstructs a vertical structure of clouds andhydrometeors up to mid-troposphere and above.In Figure 5 we show the 24-hour accumulated precipitations predicted by theMeso-NH model for the 5th of November. Starting time of the simulation is 00 UTC5 November 1994. Forecasts initialised one and two days in advance do not provideany substantial difference to the rainfall map presented here. We compared the 24-hour rainfall amounts recorded by the six rain-gauges located in northern Piedmont(shown in Figure 5 with the coloured circles) with the prediction in the grid pointsclosest to each rain-gauge. We found that the mean error, often referred to as additivebias, is about 27 mm , whereas the root mean square error is about 85 mm . eforecasting the 1994 Piedmont flood 9 Fig. 3
We note that maximum rainfall forecasts on the Alps seem to be unrealistic withvalues up to 500 mm or more in correspondence with mountain peaks.In Figures 6 and 7, we show the IVT maps for the four 6-hour steps of the 5thand 6th of November respectively. On the 5th of November (Figure 6), an elon-gated band of moisture transport connects the coasts of northern Africa to the BritishIsles, involving part of the Italian peninsula and France. This band is about 2100 kmlong and 500 km wide. Elevated IVT values, greater than the critical threshold of250 kg m − s − , are found in the area of interest, which is indicated in the plots with ared rectangle. On the 6th of November, this moist corridor is split in two parts: one tothe north of the Alps and one to the south. However, we can still observe high valuesof IVT ( ≥ kg m − s − ) in the north-eastern part of the Piedmont area, at least inthe early part of the day (i.e. until 06 UTC). Following the intensity scale proposedby Ralph et al. (2019), we note that for the Piedmont region this yields the following:(i) AR conditions (i.e. IV T ≥ kg m − s − ) lasted for at least 24 hours (see themaps in Figures 6b,c,d and 7a) and (ii) maximum instantaneous IVT values exceeded600 kg m − s − on the 5th of November (plot not shown). We can thus classify theAR that involved the P94 case as category 2 out of 5 categories following Ralph et al.(2019). Fig. 4
Vertical cross section of the Meso-NH forecast valid at 13 UTC of the 4th of November. In panels(b)-(d), wind vectors overlap the mixing ratios of three classes of hydrometeors: (b) cloud water, (c) rainand (d) graupel. Contour intervals indicate vertical velocities greater than 1 ms − . The transect is in panel(a). On the X-axis of panels (b)-(d) is indicated the distance (in m ) from the starting point ( ◦ symbol inpanel (a)) of the transect whereas on the Y-axis the altitude in m is indicated. The model orography isdashed To mark the twenty-fifth anniversary of the Piedmont flooding, we revisited the P94case by applying cutting-edge regional NWP modelling. High-resolution simulationswere forced by initial and boundary conditions produced ad-hoc with a recent ver-sion of the ECMWF-IFS global model. The goal of the work was twofold: firstly, wewanted to investigate whether a convection-permitting model is able to reconstructthe pre-frontal convection activity observed on the 4th of November in southern Pied- eforecasting the 1994 Piedmont flood 11
Fig. 5 mont. Secondly, as regards the orographic precipitation on the 5th of November in theAlpine area, we wanted to assess any new insights of a high-resolution simulation.In fact, previous studies used numerical weather models, often hydrostatic, with theparametrization of convection processes and mesh sizes in the order of few tens ofkilometres.As regards the convective precipitation in southern Piedmont on the 4th of Novem-ber, the Meso-NH forecast reconstructs the dynamics of the event well, apart from thefact that the maximum value seems to be misplaced by a few tens of kilometres tothe east of the actual position. What is important to stress is that convective cells arereconstructed in the lee side of coastal mountains in southern Piedmont, as can beappreciated by looking at the vertical cross sections in Figure 4. Using a convection-permitting model, as suggested by Buzzi et al. (1998), was likely the key to achievingsuch a result. This is an improvement with respect to previous studies (Buzzi et al.1998; Romero et al. 1998; Ferretti et al. 2000; Cassardo et al. 2002), which lackedthe reconstruction of precipitation patterns on both the north and south sides of thecoastal mountains.As regards the orographic precipitation in northern Piedmont on the 5th of Novem-ber, the high-resolution and convection-permitting Meso-NH forecast does not addany new insights with respect to past numerical investigations. As found in previousstudies, there is a fairly good consistency between model predictions and rain-gaugesdata, although the statistical analysis shown here might not be representative due tothe scarce number of observations used. Large amounts of rainfall are predicted in
Fig. 6
Vertically integrated horizontal water vapour transport (IVT) values expressed in kg m − s − for thefour 6-hour time steps on the 5th of November 1994. Wind vectors at 850-hPa isobaric level are plottedonly where IV T ≥ kg m − s − . ERA5 data were used to produce the plots. The Piedmont region ismarked with the red box the Alps in correspondence to mountain peaks, which seem to be unrealistic giventhe observations available (see data in Lionetti 1996 and Cassardo et al. 2002). Theoverestimation of the model might be due to the implementation of the ICE3 mi-crophysics scheme, which produces an excess of precipitating graupels or other hy-drometeors. Rotunno and Ferretti (2001) suggested that the simpler Kessler scheme(Kessler 1969) contains the essential set of water categories relevant for the P94 case.This scheme takes into account three type of hydrometeors: vapour, cloud water andrain. However, a test Meso-NH simulation using the Kessler scheme provided resultsthat do not differ very much from the ones presented here (map not shown). We givethe results with the ICE3 scheme, because it is the most commonly used one-momentscheme in Meso-NH (Lac et al. 2018) and thus the results obtained can be comparedwith similar works or case studies. Moreover it is more appropriate to describe thehydrometeors involved in the convective precipitation on the 4th of November asdemonstrated by the vertical cross sections shown in Figure 4.We conclude the discussion on the Meso-NH simulations by speculating why longerforecasts than the ones presented here provide results that do not differ substantiallyfrom those shown in Figures 3 and 5. We guess that one of the reasons lies in theimprovements gained over the last 25 years by global NWP modelling. Indeed, in thepaper by Petroliagis et al. (1996), the authors found good results regarding the pre- eforecasting the 1994 Piedmont flood 13 Fig. 7
As Figure 6 but for the 6th of November 1994 dictability of the P94 case. They found consistency (i.e. no sudden changes) amongconsecutive forecasts. The global analyses and forecasts we used to drive the Meso-NH simulations benefit from the improvements over the last years in terms of modelcycle and data assimilation method. Such improvements further reduce any jumpi-ness as compared to that present in the data of Petroliagis et al. (1996). It follows thatany subsequent medium-range (i.e. lead-times ≤
72 hours) regional model forecastis as reliable as a short-term (i.e. lead-times ≤
24 hours) one.The IVT maps shown in Figures 6 and 7 demonstrate that an AR landfall occurredduring the P4 case and that it played an important role in controlling the storm-total precipitation in northern Piedmont. Following the scale proposed by Ralph et al.(2019), such an AR is in category 2 out of 5. In terms of hazardous impacts, a cate-gory 2 corresponds to “Mostly beneficial, but also hazardous” (Ralph et al. 2019). Westress the fact that this subjective assessment was designed for the west coast of theUnited States and that, as mentioned in the review article by Gimeno et al. (2014), forareas with complex topography, AR landfalls are often associated with large amountsof rainfall. What we can deduce from the presence of an AR during the P94 case, isthat it supplied the necessary contribution of moisture able to change the Froude num-ber in the Alps (which is about 0.90 according to Doswell III et al. 1998). This con-clusion agrees well with the findings in Buzzi et al. (1998) and Cassardo et al. (2002),who demonstrated that the contribution of moisture was not due to the evaporationof air from the Mediterranean Sea. Furthermore, because ARs involve processes atthe meso α and β scales, this partially explains why the numerical investigations per- formed more than 25 years ago (listed in Table 1) were quite skilful in predictingthe precipitation in the mountainous part of the region. One may argue that diversealgorithms for the detection of ARs take into account not only the IVT intensity andduration (as in Ralph et al. 2019), but also criteria applied on Integrated Water Vapour(IWV) and on the geographical extent of the area where IWV conditions are met. Forinstance, Ralph et al. (2004) and Neiman et al. (2008) used an objective identificationalgorithm that involves the imposition of three conditions: (i) IWV exceeds 20 mm ,(ii) wind speeds in the lowest 2 km greater than 12.5 ms − and (iii) a long ( > km ) and narrow ( < km ) shape. For the P94 case, the conditions on the IWVand wind values are satisfied, whereas the condition on the extent of the area is notmet (namely its length is less than 2000 km ). Nevertheless, what we want to stress isthat the water vapour transported meridionally across the mountainous area of Pied-mont was sufficient to change the static stability of the air impinging the reliefs andthus favouring large amounts of rainfall. Indeed, Figure 6 confirms the presence ofa strong horizontal humidity gradient in the low-level airstreams affecting the Italianpeninsula (Rotunno and Ferretti 2001). The western part of the flow is moist, whereasthe eastern one is drier. This latter is deflected westward (i.e. to the left) over the PoValley, following the mechanism of the barrier winds (Buzzi et al. 2014; Buzzi et al.2020). Such deflection causes a convergence with the flow over Piedmont, which isvery humid. The moist flow is forced to rise over the drier one and upstream of theorographic barrier, reducing dynamically the height of the mountains to surmount. Inother words, the lower the orographic barrier (i.e. the value of h in Equation 1), thegreater the Froude number.The flooding of the Piedmont region in November 1994 is remembered becauseof its memorable impacts on the territory and the society. It drew the attention ofthe scientific community due to the atmospheric processes involved, which makethis event a testbed to investigate orographic precipitation mechanisms. Past numer-ical experiments were able to capture most of the features of the event. However,we think that reforecasting past severe weather events by testing new models (e.g.convection-permitting), parametrization (e.g. advanced microphysical schemes) anddata (e.g reanalyses) is valuable to better understand the forecasting capabilities ofcurrent systems. Acknowledgements
The Meso-NH model is freely available under the CeCILL license agreement. Theauthor wishes to thank the model’s developers and the User Support for their help.The Copernicus Climate Change Service (C3S) is acknowledged for the ERA5 data, which were used toproduce the maps in Figures 6 and 7.
Conflict of interest
The author declares that he has no conflict of interest. eforecasting the 1994 Piedmont flood 15
References
Buzzi A, Foschini L (2000) Mesoscale meteorological features associated with heavyprecipitation in the southern Alpine region. Meteorology and Atmospheric Physics72(2-4):131–146Buzzi A, Tartaglione N, Malguzzi P (1998) Numerical simulations of the 1994 pied-mont flood: Role of orography and moist processes. Monthly Weather Review126(9):2369–2383Buzzi A, Davolio S, Malguzzi P, Drofa O, Mastrangelo D (2014) Heavy rainfallepisodes over Liguria in autumn 2011: numerical forecasting experiments. Nat-ural Hazards and Earth System Sciences 14(5):1325Buzzi A, Di Muzio E, Malguzzi P (2020) Barrier winds in the Italian region andeffects of moist processes. Bulletin of Atmospheric Science and Technology pp1–32Caniaux G, Redelsperger J, Lafore JP (1994) A numerical study of the stratiformregion of a fast-moving squall line. Part I: General description and water and heatbudgets. Journal of the Atmospheric Sciences 51(14):2046–2074Cassardo C, Loglisci N, Gandini D, Qian M, Niu GY, Ramieri P, Pelosini R,Longhetto A (2002) The flood of November 1994 in Piedmont, Italy: a quanti-tative analysis and simulation. Hydrological processes 16(6):1275–1299Doswell III CA, Ramis C, Romero R, Alonso S (1998) A diagnostic study of threeheavy precipitation episodes in the western Mediterranean region. Weather andForecasting 13(1):102–124Ferretti R, Low-Nan S, Rotunno R (2000) Numerical simulations of the Piedmontflood of 4–6 November 1994. Tellus A 52(2):162–180Gimeno L, Nieto R, V´azquez M, Lavers DA (2014) Atmospheric rivers: A mini-review. Frontiers in Earth Science 2:2Hagelin S, Son J, Swinbank R, McCabe A, Roberts N, Tennant W (2017) The MetOffice convective-scale ensemble, MOGREPS-UK. Quarterly Journal of the RoyalMeteorological Society 143(708):2846–2861Hersbach H, Bell W, Berrisford P, Hor´anyi A, J MS, Nicolas J, Radu R, Schepers D,Simmons A, Soci C, Dee D (2019) Global reanalysis: goodbye ERA-Interim, helloERA5. ECMWF Newsl pp 17–24Jansa A, Genoves A, Garcia-Moya JA (2000) Western Mediterranean cyclones andheavy rain. Part 1: Numerical experiment concerning the Piedmont flood case. Me-teorological Applications 7(4):323–333Kessler E (1969) On the distribution and continuity of water substance in atmosphericcirculations. In: On the distribution and continuity of water substance in atmo-spheric circulations, Springer, pp 1–84Krichak S, Feldstein SB, Alpert P, Gualdi S, Scoccimarro E, Yano J (2016) Dis-cussing the role of tropical and subtropical moisture sources in cold season extremeprecipitation events in the Mediterranean region from a climate change perspective.Natural Hazards and Earth System Sciences 16(1):269Lac C, Chaboureau JP, Masson V, Pinty JP, Tulet P, Escobar J, Leriche M, Barthe C,Aouizerats B, Augros C, et al. (2018) Overview of the Meso-NH model version5.4 and its applications. Geoscientific Model Development 11(5):1929
Lafore JP, Stein J, Asencio N, Bougeault P, Ducrocq V, Duron J, Fischer C, H´ereil P,Mascart P, Masson V, et al. (1997) The Meso-NH atmospheric simulation system.Part I: Adiabatic formulation and control simulations. In: Annales geophysicae,Springer, vol 16, pp 90–109Lavers DA, Villarini G (2013) The nexus between atmospheric rivers and extremeprecipitation across Europe. Geophysical Research Letters 40(12):3259–3264Leutbecher M, Lock SJ, Ollinaho P, Lang ST, Balsamo G, Bechtold P, Bonavita M,Christensen HM, Diamantakis M, Dutra E, et al. (2017) Stochastic representationsof model uncertainties at ECMWF: state of the art and future vision. QuarterlyJournal of the Royal Meteorological Society 143(707):2315–2339Lionetti M (1996) The Italian floods of 4–6 November 1994. Weather 51(1):18–27Lunet T, Lac C, Auguste F, Visentin F, Masson V, Escobar J (2017) Combinationof WENO and explicit Runge-Kutta methods for wind transport in the Meso-NHmodel. Monthly Weather Review 145(9):3817–3838Neiman PJ, Ralph FM, Wick GA, Lundquist JD, Dettinger MD (2008) Meteorolog-ical characteristics and overland precipitation impacts of atmospheric rivers af-fecting the West Coast of North America based on eight years of SSM/I satelliteobservations. Journal of Hydrometeorology 9(1):22–47Nickerson EC, Richard E, Rosset R, Smith DR (1986) The numerical simulation ofclouds, rains and airflow over the Vosges and Black Forest mountains: a meso- β model with parameterized microphysics. Monthly Weather Review 114(2):398–414Palmer TN (2001) A nonlinear dynamical perspective on model error: A proposal fornon-local stochastic-dynamic parametrization in weather and climate predictionmodels. Quarterly Journal of the Royal Meteorological Society 127(572):279–304Peralta C, Ben Bouall`egue Z, Theis S, Gebhardt C, Buchhold M (2012) Account-ing for initial condition uncertainties in COSMO-DE-EPS. Journal of GeophysicalResearch: Atmospheres 117(D7)Petroliagis T, Buizza R, Lanzinger A, Palmer TN (1996) Extreme rainfall predictionusing the European Centre for Medium-Range Weather Forecasts ensemble pre-diction system. Journal of Geophysical Research: Atmospheres 101(D21):26227–26236Ralph FM, Neiman PJ, Wick GA (2004) Satellite and CALJET aircraft observationsof atmospheric rivers over the eastern North Pacific Ocean during the winter of1997/98. Monthly Weather Review 132(7):1721–1745Ralph FM, Rutz JJ, Cordeira JM, Dettinger M, Anderson M, Reynolds D, SchickLJ, Smallcomb C (2019) A scale to characterize the strength and impacts of atmo-spheric rivers. Bulletin of the American Meteorological Society 100(2):269–289Raynaud L, Bouttier F (2016) Comparison of initial perturbation methods for ensem-ble prediction at convective scale. Quarterly Journal of the Royal MeteorologicalSociety 142(695):854–866Romero R, Ramis C, Alonso S, Doswell III CA, Stensrud DJ (1998) Mesoscale modelsimulations of three heavy precipitation events in the western Mediterranean re-gion. Monthly Weather Review 126(7):1859–1881Rotunno R, Ferretti R (2001) Mechanisms of intense Alpine rainfall. Journal of theAtmospheric Sciences 58(13):1732–1749 eforecasting the 1994 Piedmont flood 17eforecasting the 1994 Piedmont flood 17