1 Introduction

The infamous Piedmont flood of 1994 affected a relatively vast area in northwestern Italy, almost entirely contained in the administrative region of Piedmont (Fig. 1). Several rivers overflowed their banks, some having their catchment basins in the western Alps, where the chain exhibits its largest concavity, corresponding to northwestern Piedmont. A severe flooding affected also the Tanaro River, whose catchment is located in a different area, between the Ligurian Apennines and the Maritime Alps in the southern portion of Piedmont (see also Fig. 1 in Buzzi et al. 1998 and Fig. 3). Flooding of the former group of rivers was associated with the huge orographic rainfall maximum present on the southeastern flank of the Alpine chain as visible in Fig. 1, with accumulated maxima exceeding 500 mm in 48 h. Flooding of the Tanaro River was associated with the secondary maximum between the Alps and the Apennines, also visible in Fig. 1, with corresponding observed values of 48 h accumulated precipitation around 250 mm (Cassardo et al. 2002).

Fig. 1
figure 1

(Left) Relevant geographical details of the area of interest. “Pied”, “L” and “Cor” indicate Piedmont and Liguria regions, and Corsica, respectively. Maritime Alps (MAlp) and Ligurian Apennines (Lap) are indicated. Orography is also plotted using grey shading corresponding to 500, 1000 and 2000 m. Lines indicate region administrative borders. (Right) Total 72 h observed precipitation (mm) accumulated between 04 and 06 November 1994 over Piedmont region (courtesy of ARPA Piemonte). Black lines indicate province administrative borders; the rainfall field is obtained by interpolation of the regional rain gauge network data and plotted only over the Piedmont region area

Full size image

Although the Piedmont flood was not a flash-type flood, since it was due to rainfall insisting over the same areas for a couple of days (except for high intensity rainfall peaks of shorter duration over the Apennines—see below), it was severely underpredicted, and the alert system underwent a total failure. The disaster revealed the deficiencies in operational meteorological and hydrological monitoring and forecasting at the time. It stimulated a substantial effort, at national and, to a larger extent, regional levels, for the implementation of modern observational networks and, in a second stage, for the development of meteorological and hydrological forecasting systems based on the application of regional numerical models. On the one hand, at the time the European Centre for Medium-range Weather Forecasts (ECMWF) global model had a totally inadequate resolution (T213, corresponding to a grid distance of about 90 km, Branković and Molteni 1997) for an accurate quantitative precipitation forecasting (QPF). On the other hand, development and operational application of higher resolution limited area models in Italy were in its pioneering stage. For example, the development of the BOLAM model had started a few years before the event: the first paper presenting the model (Buzzi et al. 1994) had just been published. The very few nonhydrostatic models available at the time were used only for scientific purposes.

The Piedmont flood of 1994 became a testbed for subsequent projects in the field of dynamical meteorology and meteorological model development, testing the models capability of predict the event and, in particular, the precipitation amount, and for the coupling between meteorological and hydrological models. For example, in the context of the European projects RAPHAEL (Bacchi and Ranzi 2000) and HERA (Volkert 2000), a multimodel investigation of the flood event led to the following results: (a) using the ECMWF analyses as initial conditions, hydrostatic models with a sufficiently high resolution (grid distance of the order of 10–20 km) were capable of a reasonable forecast of the orographic precipitation over the Alps, while (b) the precipitation maximum over the Apennines/Maritime Alps remained underpredicted. In particular, models tended to predict it south of the orographic divide, so that the associated Tanaro River flood could not be adequately represented by hydrological models tested in cascade (Buzzi et al. 1998; Buzzi and Foschini 2000).

A deeper analysis of meteorological model results, and of the causes of their deficiencies, allowed distinguishing the mechanisms associated with the two main precipitation maxima. The main maximum in the western Alps was attributed to orographic lifting in quasi-neutral stratification (taking into account the reduction of static stability in saturated conditions—see Lalas and Einaudi 1974), so convection in this area did not play a basic role. Conversely, multiple convective cells originating over the Ligurian Sea and travelling to the north (Buzzi and Foschini 2000) contributed essentially to the southerly maximum. Consequently, the underforecasting of precipitation amount in the Tanaro River catchment could be attributed to the limited capability of convective parameterization schemes to adequately represent convective precipitation in hydrostatic models. In addition to the latter aspect, it was realized that hydrostatic dynamics (papers quoted above) in the presence of flow over a ridge, as in the case of the Ligurian topography, tend to produce larger amplitude descent associated with the lee wave dynamics. This means that hydrostatic models tend to confine precipitation to the upstream side of a ridge, suppressing it at the divide and downstream. Another model aspect that was found to contribute to the QPF error over the northern side of the Apennines was the lack of advection of falling precipitation (snow in particular), considering that, in cases of strong winds, precipitation can fall tens of km downstream of the area of formation aloft. Most of the regional models, including BOLAM, implemented advection of hydrometeors at the end of 1990s (see again Buzzi and Foschini 2000; Ferretti et al. 2000).

During the 1990s, a major international research experiment, the Mesoscale Alpine Programme (MAP; Bougeault et al. 2001) was planned, culminating in the observational field phase in 1999. The study of orographic “dry” and “wet” processes associated with orography, addressed to the improvement of weather forecasting in mountainous regions, was the main objective of MAP. The 1994 Piedmont flood event was selected as one of the most significant events in the Alps to be studied, in order to better understand the dynamical and microphysical processes responsible for orographic heavy precipitation, and to improve meteorological models used both in research and forecasting. Some studies of the MAP stage preceding the field campaign led to the formulation of conceptual models that were subsequently confirmed and refined, after the MAP field phase results became available. Among them, Buzzi et al. (1998), by comparing numerical experiments designed to reveal the effects of moist processes on mesoscale dynamics, showed the importance of equivalent (moist) static stability in governing the establishment of flow-over versus flow-around regimes. Such processes, implying very important feedbacks between dynamics and “microphysics” (i.e., moist processes associated with atmospheric condensation and evaporation), were the object of several subsequent studies based on the interpretation and modelling of meteorological events of heavy precipitation in the Alpine area, observed during the MAP field phase (see, for example, Rotunno and Ferretti 2001; Rotunno and Ferretti 2003; Bousquet and Smull 2003; Chiao et al. 2004; Lascaux et al. 2006; Richard et al. 2007; Rotunno and Houze 2007).

The topic of heavy precipitation and floods in the Mediterranean area has continued to be of great interest for the scientific community, and it has been the core of the research activities of the Hydrological Cycle in the Mediterranean Experiment (HyMeX, Drobinski et al. 2014) during the last decade. The field campaign in fall 2012 (Ducrocq et al. 2014) was entirely devoted to study the formation of quasi-stationary mesoscale convective systems and their interaction with the complex orography surrounding the basin, in order to advance the scientific knowledge and eventually improve forecasting. Compared with MAP, higher resolution models and improved monitoring facilities and scientific instrumentation allowed investigating the small scales of convection, providing new insights on the physical processes (e.g., Davolio et al. 2016; Pichelli et al. 2017; Lee et al. 2018).

This paper is also an opportunity to provide an overview of the research activity in the field of numerical modelling at the Institute of Atmospheric Sciences and Climate (ISAC) of the National Research Council of Italy (CNR), of its main applications and current operational implementation. The general synoptic situation leading to this severe event is described in the “Weather evolution” section. The ISAC meteorological suite is presented in the “Overview of ISAC models” section. The “Simulation results on different space and time scales” section is devoted to the presentation of the main results achieved by both the monthly ensemble forecasts and the short-term simulations. The “Convection over the Ligurian Sea and related mesoscale phenomena” section describes physical mechanisms responsible for convective initiation during the first phase of the event, and the “Conclusions” section provides summary and concluding remarks.

2 Weather evolution

The synoptic situation between 4 and 6 November 1994 (Fig. 2) was typical of Alpine heavy precipitation events in autumn, as identified and described in many papers (see, for example, Buzzi et al. 1998; Massacand et al. 1998; Ferretti et al. 2000; Buzzi et al. 2014; Grazzini et al. 2020): a deep trough over the eastern Atlantic approached the European coast and then deepened over the western Mediterranean basin. The slow eastward progression of the trough and its increasing amplitude, associated with a positive potential vorticity anomaly (streamer) in the upper troposphere, favoured an intensification of geopotential gradient on the eastern side of the trough and, in turn, of the moisture transport towards the Alps, in the form of a prefrontal southerly low-level jet develo** over the Tyrrhenian Sea.

Fig. 2
figure 2

ECMWF ERA5 reanalysis valid at 1200 UTC, 04 November 1994: 500 hPa Geopotential height (colour shading) and mean sea level pressure (white isolines)

Full size image

Several processes contributing to the event were identified in previous papers: Ferretti et al. (2000) showed that rising motion associated with synoptic-scale ascent was further enhanced by orographic lifting of conditionally unstable air and by latent heat release. Positive vorticity due to vortex stretching produced an easterly perturbation in the wind field over the Po Valley, which slowed the eastward progression of the precipitation system, and induced further lifting of unstable air. Buzzi et al. (1998) identified the critical role of latent heat conversions, associated with moist processes, for the formation of a multiple front–like rainband structure and for determining the orographic flow regime.

Two areas of northwestern Italy were hit by the flood event. Heavy rainfall characterized by intense convective activity was recorder especially over the area where the Maritime Alps join the Ligurian Apennines (hereafter referred to as Apennines) since the afternoon of 4 November (Fig. 1). Rainfall affected both sides of the orography divide, the upwind Liguria catchments as well as the downwind drainage areas of the Po river tributaries (the most important being the Tanaro River, Fig. 3) in southern Piedmont. Later on, during 5 November, a second and more intense precipitation spell affected also the northwestern Alps especially on the upslope area. This spell was mainly ascribed to intense orographic ascent of the prefrontal low-level jet (Buzzi et al. 1998; Cassardo et al. 2002), possibly enhanced by mesoscale mechanisms as described above. Precipitation intensity weakened in the morning of 6 November. Despite the total accumulated precipitation was higher over the Alps, exceeding 350 mm in 36 h on a wide area, locally larger than 500 mm, the major floods occurred in southern Piedmont, within watersheds along the northern side of the Apennines. In fact, precipitation was uniformly distributed during the 2 days over the Alps, while it was mostly concentrated in the night between 4 and 5 November over the Apennines. Extensive damages to agriculture, infrastructures and private properties, as well as 70 casualties, were the dramatic, final result of this destructive flood.

Fig. 3
figure 3

(Left) Operational integration domain of BOLAM (full image) and MOLOCH (dashed box). (Right) Integration domain for the high-resolution (500 m grid spacing) MOLOCH experiment and indication of the rivers Po and Tanaro (Tan). Rain gauge locations: C = Capanne di Cosola, P = Ponzone, R = Priero, L = Lanzo, E = Eugliano and O = Oropa. In both panels, model orography is also plotted using grey shading corresponding to 500, 1000 and 2000 m

Full size image

3 Overview of ISAC models

3.1 Historical background and applications

A systematic activity in develo** original atmospheric models at ISAC was initiated in the early 1990s. A limited area hydrostatic model (BOLAM, Buzzi et al. 1994) was set up mainly with the purpose of providing a scientific and operational tool for forecasting severe meteorological phenomena like heavy precipitation and strong winds over Europe and more specifically over the Mediterranean area. In fact, in that period, after the occurrence of major floods, it became clear also in Italy that limited area models could provide much more accurate short-range forecasts, especially of precipitation, than the global models that were almost the exclusive source of information at the time.

More recently, a non-hydrostatic limited area model (MOLOCH, Malguzzi et al. 2006) was developed as a tool for high resolution (1–3 km grid spacing) forecasting, allowing the explicit treatment of atmospheric convection. Finally, a global atmospheric model (GLOBO, Malguzzi et al. 2011) was also developed at ISAC, based mainly on the equations and parameterization set of BOLAM, using a simplified ocean and sea ice scheme. More recently, BOLAM and GLOBO have been unified in a single code.

The above models are suitable to simulate and predict the atmospheric circulation from the global to the local scale, at different temporal ranges, with high computational efficiency. They have been used for numerous scientific studies and applications, as for instance: sensitivity and impact studies, and diagnostics of meteorological phenomena including severe weather and storms (e.g. Malguzzi et al. 2006; Cavaleri et al. 2010; Fantini et al. 2012; Cioni et al. 2016; Davolio et al. 2016, 2017a; Buzzi et al. 2020); model coupling with hydrological and ocean models (Davolio et al. 2015; Lombardi et al. 2018; Ferrarin et al. 2013, 2019; Poletti et al. 2019); theoretical and idealized studies of instability processes (Davolio et al. 2009; Fantini and Malguzzi 2008); applications to probabilistic and ensemble forecasting, and atmospheric predictability studies (Uboldi and Trevisan 2015; Corazza et al. 2018); data assimilation studies (Tiesi et al. 2016; Davolio et al. 2017b); model validation and intercomparison projects (Nagata et al. 2001; Casaioli et al. 2013). The ISAC NWP models have also been employed as basic tools in many international field experiments such as the international forecasting demonstration project called MAP D-PHASE (Rotach et al. 2009) and the HyMeX campaign SOP1 (Ducrocq et al. 2014; Ferretti et al. 2014), and in numerous European scientific projects.

Besides the scientific applications, ISAC models, specifically BOLAM and MOLOCH, are used for operational forecasting practice by several Italian and foreign institutions (e.g. Lagouvardos et al. 2003; Mariani et al. 2015; Corazza et al. 2018). GLOBO is among the models that provide extended range forecasts on the monthly time scale in the context of the Subseasonal-to-Seasonal (S2S) project under the WWRP/WCRP program (Vitart et al. 11, in which the typical pattern of a V-shape mesoscale convective system over the sea, affecting the Ligurian region, can be clearly identified, together with an intense convective activity developed over the northern tip of the Corsica Island.

Fig. 11
figure 11

NOAA-AVHRR IR images at 1941 UTC, 04 November 1994. The red circle indicates the V-shape MCS described in the text. The shape of the Italian peninsula is recognizable below the label “Italy”

Full size image

All the numerical experiments generate a low-level wind convergence line just offshore the Liguria coastline, in a position suitable to explain the presence of the observed mesoscale convective system and compatible with the location of intense rainfall affecting the western part of the region. In fact, the forecast presented in Fig. 12 shows intense precipitation from the Apennines divide to the coast. The shape of the rainfall pattern, elongated over the sea, reveals the presence of the convergence line in the low levels. The convergence line is evident in Fig. 12, showing the 10-m wind field at 2100 UTC, 4 November, simulated by the MOL12 experiment.

Fig. 12
figure 12

(left) 12 h accumulated precipitation (experiment MOL12) between 1200 UTC, 04 November, and 0000 UTC, 05 November 1994 (colour bar in mm); (right) 10 m wind (colour bar in ms−1) at 2100 UTC, 04 November 1994

Full size image

Once the event proceeds, the approaching cold front determines an intensification of the southerly flows, so that the equilibrium along the convergence line is disrupted and the inflow from the sea is able to overcome the Apennines barrier, reaching the Alps.

6 Conclusions

The 25th anniversary of the infamous Piedmont flood of 1994 has been an opportunity to apply the modelling capabilities developed at CNR-ISAC to a high-impact meteorological and hydrological event that stimulated scientific research in the field of dynamic meteorology, numerical modelling and forecasting. Reexamining this extreme event in light of the current scientific knowledge and numerical modelling resources allows confirming and even deepening the results obtained about 20 years ago, and provides useful information for the current and future operational implementation of ISAC NWP models.

Simulation results based on multiple modelling tools interestingly indicate that this event is characterized by a relatively large predictability that goes, to some extent, into the subseasonal range. Moreover, they confirm the higher predictability of precipitation over the alpine slopes of Piedmont, while rainfall prediction over the Apennines, between Piedmont and Liguria, turns out to be more challenging due to the intense convective activity affecting this area, especially during the first phase of the event.

Convection-permitting model simulations successfully describe the mesoscale mechanisms leading to the formation and regeneration of convective systems offshore the Liguria coast, as well as their propagation to the north, leading to heavy precipitation over the Apennines divide, partly extending to the lee side areas corresponding to the Tanaro River catchment. These simulations, though still more or less underestimating precipitation peaks, provide a much more accurate rainfall prediction than up-do-date hydrostatic models, and demonstrate the capability to reproduce the observed very intense precipitation maxima at very high spatial resolution, i.e. in the subkilometre range. Most of the earlier hypotheses formulated about the possible causes of model errors (like limitations inherent in hydrostatic dynamics and of convective parameterization schemes, insufficient resolution in the presence of complex topography, simplified microphysics not describing advection of hydrometeors) have been confirmed after models have become by far more accurate. However, we do not intend to underestimate the importance of data assimilation and formulation of initial conditions: these crucial aspects of NWP would require a specific treatment and are out of the scope of the present paper.

For what concerns possible practical indications suggested by this study, the recently adopted operational setup based entirely on ISAC models, namely the application in cascade of GLOBO, BOLAM and MOLOCH, is somehow supported by the results of this case study, but may need further investigation. Moreover, even the direct nesting of MOLOCH into the global IFS analysis and forecast fields seems feasible since it is not affected by relevant noise propagation, neither due to the initial condition nor at the boundaries. However, when implementing a modelling chain over an area such as the Mediterranean basin, and in particular Italy that is characterized by complex orography and sea-land transition, the choice of the integration domain is critical. These simulation results, as well as the expertise after many years of NWP practise, suggest that this aspect is at least as important as model resolution, and thus it requires particular attention.