An Extensive Italian Database of River Embankment Breaches and Damages

Abstract

River embankments are critical flood defense structures, stretching for thousands of kilometers across alluvial plains. They often originated as natural levees resulting from overbank flows and were later enlarged using locally available soils yet rarely designed according to modern engineering standards. Substantially under-characterized, their performance to extreme events provides an invaluable opportunity to highlight their vulnerability and then to improve monitoring, management, and reinforcement strategies. In May 2023, two extreme meteorological events hit the Emilia-Romagna region in rapid succession, causing numerous breaches along river embankments and therefore widespread flooding of cities and territories. These were followed by two additional intense events in September and October 2024, marking an unprecedented frequency of extreme precipitation episodes in the history of the region. This study presents the methodology adopted to create a regional database of 66 major breaches and damages that occurred during May 2023 extensive floods. The database integrates multi-source information, including field surveys; remote sensing data; and eyewitness documentation collected before, during, and after the events. Preliminary interpretation enabled the identification of the most likely failure mechanisms—primarily external erosion, internal erosion, and slope instability—often acting in combination. The database, unprecedented in Italy and with few parallels worldwide, also supported a statistical analysis of breach widths in relation to failure mechanisms, crucial for improving flood hazard models, which often rely on generalized assumptions about breach development. By offering insights into the real-scale behavior of a regional river defense system, the dataset provides an important tool to support river embankments risk assessment and future resilience strategies.

1. Introduction

Floods are the most impactful natural disasters in Europe, with 3563 events recorded between 1980 and 2015 [1]. They account for the highest economic losses and are second only to extreme temperatures in terms of fatalities [2]. The situation in developing countries is even more severe. Asia is the most flood-affected continent, accounting for 40% of global flood disasters, 65% of the people killed, and 96% of those affected [3,4,5]. According to a UNESCO study [6], floods in Asia have caused an average of 22,800 deaths per year and economic losses estimated at EUR 125 billion—impacts exacerbated by a flood frequency that has doubled over the past 30 years [4]. In 2010, nearly 1 billion people worldwide lived in flood-prone areas—a number expected to exceed 1.6 billion by 2050. Urbanization exacerbates exposure, with 70% of the global population projected to live in cities, many of which are expanding into flood-susceptible zones [7]. Between 1980 and 2013, river floods caused globally over EUR 852 billion in damages and more than 220,000 fatalities [8]. In Europe, annual economic losses are expected to increase fivefold by 2050, with even steeper rises forecast for Asia and Sub-Saharan Africa [3,7]. Climate change is expected to intensify flood risks due to more frequent and severe precipitation events, highlighting the need for robust flood risk assessment and adaptive mitigation strategies [9,10,11].

In Italy, flood risk is a widespread concern, affecting a significant share of the population and territory. According to ISPRA [12], 5.4% of the national land area is exposed to high-probability flood events (HPH—High Probability Hazard—return periods of 20–50 years), impacting 3.5% of the population (2.43 million people). This exposure increases to 10% and 14% of the land area, and over 11% and 20% of the population for medium (MPH—return period of 100–200 years) and low (LPH—return period > 200 years) probability flood scenarios, respectively (Figure 1A,B).

The Emilia-Romagna region (44°29′38″ N, 11°20′34″ E), in Northern Italy (Figure 2A), is historically the most vulnerable region to hydrogeological hazards, such as floods and landslides [12]. It holds the national record for flood-prone territory, with 45% of its area and over 60% of its population exposed under medium (M) and low (L) probability flood scenarios (Figure 1A,B, red boxes). The high flood risk results from a combination of factors: land morphology, configuration of the hydraulic system, and economic and demographic characteristics of the region. Emilia-Romagna is a key area for Italian industry and agriculture, with high urbanization, dense population (Figure 1C–E), and major tourist destinations, such as Bologna, Parma, Modena, Ravenna, and the Adriatic coast (e.g., Rimini and Riccione).

Between May 2023 and October 2024, the region was hit by four extreme rainfall events that caused severe damage to infrastructures, agricultural and industrial activities, residential areas, and hydraulic structures, with impacts extending from Apennine hills to the Po Valley plains (Figure 2A).

The first two events, which occurred in quick succession in May 2023 (1–3 May and 16–17 May), were the most severe. Their exceptional nature lies not only in their spatial extent and persistence, causing the highest number of breaches and flooded areas, some of which are illustrated in Figure 3, but also in their rarity. The mean recurrence interval of two such extreme events occurring so close in time is estimated to be in the range of several hundred years. The second event surpassed even the historically significant 1939 flood, making the 2023 floods unprecedented in regional records [15,16]. The documented consequences were dramatic [16]:

• Simultaneous overflow of 23 rivers, with an estimated total flooded volume of 350 million m³ and approximately 540 km² of flooded territory;
• 65,598 landslides, affecting over 72 km² of territory;
• Damage to 1950 road infrastructures, severely disrupting regional mobility;
• 17 fatalities and more than 10,000 displaced people;
• Estimated economic losses amount to ~EUR 8.8 billion [17].

These events served as a stark reminder of the region’s vulnerability to hydrogeological risks in the era of climate change and of the urgent need for comprehensive and systemic mitigation strategies. Understanding how key components of the regional infrastructure—particularly river embankments—respond to such extreme events is essential.

This study addresses this issue by analyzing river embankment failures during the May 2023 floods, which effectively provided a real-scale test of the regional river defense system. Data were gathered from multiple sources, including video, photographic, and satellite imagery, as well as post-event inspections and consultations with civil protection agencies and river basin authorities. All relevant data were systematized into a regional database. Globally, increasing attention has been given to breach documentation, leading to the development of various river embankment performance databases. Among the most extended are the U.S. Army Corps of Engineers’ National Levee Database [18], which lists over 10,000 failures, and the International Levee Performance Database (ILPD), which includes 1500 detailed cases [1] and the Italian Sand Boil Database [19]. However, most are not open-access or lack detailed geotechnical and hydraulic data, or they group together heterogeneous cases from diverse settings, reducing their comparability.

The Emilia-Romagna embankment breach database is notable for its exceptional nature—marked by the broad geographical extent of the event, the consistency of the recorded data, and the unusually high number of levee breaches, including 66 major failures. The contents include the following: (a) systematic identification and localization of major failures and damages; (b) preliminary interpretation and statistical classification of failure mechanisms; (c) statistical analysis of breach dimensions, particularly breach widths by failure type. These analyses can support large-scale assessment of more vulnerable embankment sections, guide risk mitigation strategies, provide a foundation for future investigations aimed at identifying recurrent vulnerabilities, refining breach modelling approaches, and supporting targeted mitigation and resilience strategies.

2. Study Area and Meteorological Extreme Events

2.1. The Exceptional Meteorological Events of May 2023 in Emilia Romagna

In the spring of 2023, Emilia-Romagna was hit by two severe weather events in rapid succession: the first from 1 to 3 May and the second from 15 to 17 May. These primarily affected the eastern portion of the Po River District (Figure 2A) and are hereafter referred to as the “first event” and the “second event”, respectively. Both were triggered by slow-moving low-pressure systems centered over central Italy, arriving about two weeks apart, following an unusually dry period (January–April 2023) marked by a significant rainfall deficit, resulting in notably reduced soil moisture across the Po Plain. A comprehensive regional-scale meteorological analysis has been provided by Arpae (Regional Agency for Prevention, Environment and Energy) in reports [15,20].

The first event mainly affected the provinces of Forlì-Cesena, Ravenna, Bologna, and Modena (Figure 2A), with continuous weak to moderate rainfall from the night of 1 May through the morning of 3 May. Cumulative rainfall was particularly intense on 2 May, exceeding 200 mm in the central-eastern sector of the region, from Bologna to the Adriatic coast (Figure 4A, purple areas). The western part, except for the Modena province, saw lower, though still significant, accumulations below 100 mm. On 3 May, the event weakened, with daily totals dropping below 50 mm across the region. Despite moderate intensities compared to past events, its duration and wide spatial extent made it exceptional. It was the most intense 48 h rainfall event in Emilia-Romagna since 1997 and the wettest spring event since 1961 [20]. On 4 May, a state of emergency was declared for six provinces (Reggio Emilia, Modena, Bologna, Ferrara, Ravenna, and Forlì-Cesena), covering about 15,500 km² (5.1% of the national territory).

The second event began on 16 May, with a Mediterranean weather system bringing widespread and intense rain to areas already affected by the first event. On 17 May, the intensity peaked in the provinces of Ferrara, Bologna, and Modena before gradually decreasing and moving out of the region by midday. Total rainfall on 16 May exceeded 100 mm, breaking historical records at several stations. On 17 May, values were lower as the event faded. The most affected provinces were Forlì-Cesena, Ravenna, Bologna, and Rimini (Figure 2A), three of which had also been heavily impacted during the first event.

Over 16–17 May, cumulative rainfall reached 200 mm, with peaks over 240 mm in the hilly areas of Forlì-Cesena and Ravenna (Figure 4B, purple areas), near the Tuscany border [15]. In just 36 h, an estimated 350 million cubic meters of water fell—equivalent to the region’s average six-month rainfall [21,22]. Comparing Figure 4A,B reveals that the epicenter of rainfall remained largely unchanged for both events, though maximum totals were higher in the second event in relation to the first event (around 240 mm and 210 mm, respectively). Following the second event, the emergency status was extended to include Rimini province, part of the Pesaro and Urbino area in the Marche region, and municipalities within the Metropolitan City of Florence in Tuscany (Figure 1B). The affected territory increased to 17,300 km², about 5.7% of the national total.

2.2. Effects of the Precipitation Events on the Hydrographical Basins

The Emilia-Romagna region is subdivided into four hydrographical basins for water resource management: the Po Basin, the Reno Basin, the Regionale-Romagnolo Basin, and the Conca-Marecchia Basin. This subdivision reflects the natural water drainage pattern of the region and takes into account the geographical and environmental features of each basin [23]. Figure 5 shows the geographical extent of the basins. As shown in Figure 6A, the cumulative precipitation from the first meteorological event exceeded 150 mm in several hilly areas of the Reno Basin (Samoggia, Idice, Quaderna—tributary of Idice—Sillaro, Santerno, and Senio rivers) and of the Regionale-Romagnolo Basin (particularly the Lamone and Montone rivers). The rivers experienced simultaneous peaks in hydrometric levels, many of which exceeded historical maxima—values that vary depending on the river and gauging station [20].

Lower cumulative precipitations (~100 mm) were recorded in the hilly areas of the western part of the Po Basin (Crostolo, Secchia, Panaro and their tributaries) and of the Reno Basin (Reno tributaries, from Reno’s headwater to Savena Abbandonato). In these areas, hydrometric levels did not exceed historical records, indicating less critical conditions in comparison to the central and eastern sectors [20]. Figure 6B shows that cumulative precipitation during the second event also exceeded 150 mm in the hilly areas of the Reno Basin (Samoggia, Idice, Sillaro, Santerno Senio rivers) and of the Regionale-Romagnolo Basin (Lamone, Montone, Ronco, Savio, Bevano, and Rubicone rivers). Hydrometric levels once again exceeded historical maxima—many of which had just been set during the first event—across nearly all rivers from the Samoggia (western Reno Basin) to the Marecchia (Conca-Marecchia Basin). In the western part of the Po Basin (Secchia and Panaro rivers and tributaries) and along the Reno River, the flooding was less severe, though in many cases water levels approached the embankment crest and surpassed the warning thresholds.

Overall, the second event was more intense and spatially extensive than the first. Its impact on the hydrological system was amplified by already saturated soil conditions, resulting in higher and more critical hydrometric levels.

The cumulative impact of both events led to the flooding of approximately 540 km² of regional territory (Figure 7). Flood mapping was conducted using remote sensing imagery and digital photos and videos collected via helicopter and drone surveys photos and videos helicopter and drone surveys. These results are available through the Emilia-Romagna Geoportal [24].

It is relevant to highlight that reconstruction of the flood extent is affected by inherent uncertainties, particularly in this case due to the simultaneous contribution of multiple factors: overtopping or breaching of embanked rivers or secondary channels, urban flooding due to drainage system overload, saturation-induced runoff in rural areas, and rising groundwater levels above the ground surface [23]. Brath et al. [16] analyzed the return period of the maximum daily aerial mean rainfall during the May 2023 events for each hydraulic basin (Figure 8). Using the Generalized Extreme Value (GEV) probability distribution on historical data (1921–present), they found extreme return periods: >500 years for the Senio, Lamone, and Montone rivers, and 340 years for Ronco (Figure 8). In addition to flooding, the events triggered widespread landslides. A total of 65,598 landslides were recorded, covering an area of 72.21 km² [16]. Of this area, 62% is located in the hilly-mountainous zone of the Forli-Cesena province (Montone and Lamone Basins) and 17% in the Ravenna province (Senio Basin), the sectors that also received the highest cumulative rainfall, as indicated by the isolines in Figure 6. Interactive maps and data visualizations are available on the Emilia-Romagna Geoportal [25].

2.3. The River Embankment System of the Region

The earliest documented reclamation in the Po Plain dates back to the Roman era [26]. As communities undertook large-scale land modifications to address the sanitary and hydraulic challenges of the marshy terrain. These interventions aimed to support agriculture and the establishment of settlements through extensive reclamation works. During Roman colonization, river channels downstream of the Via Emilia were poorly defined and prone to frequent changes due to floodings [27]. These conditions challenged the Roman centuriation system, which nevertheless extended into coastal marshlands. It was during this period that the first river embankments were constructed to confine watercourses and prevent their avulsion and flooding. More systematic management practices began to emerge only after the 11th century.

Local chronicles describe sporadic efforts to stabilize riverbeds and control floods, especially during the resurgence of urban life and agriculture in formerly centuriated lands. These interventions aimed to progressively reclaim lands and establish stable conditions for cultivation and habitation. In the 19th century, marshlands-related issues were more comprehensively addressed with the introduction of canal systems for industrial water use [28]. Human interventions have since profoundly shaped the hydrography of the Po Valley [26], which exhibits marked differences between its western and eastern sectors. In the west, rivers follow relatively parallel paths, crossing valleys and plains, before flowing into the Po River. In the east, starting with the Reno basin, rivers discharge directly into the Adriatic Sea.

Today’s river embankments are the result of centuries of progressive transformations, carried out with varying construction techniques and materials. The core of the embankment is typically natural, composed of coarser component of the sediments deposited near riverbeds by successive floods (natural river embankment). Over time, embankments have been artificially raised and reinforced, often using locally available materials from the adjacent fields. In mountainous and hilly areas, rivers are generally confined by natural slopes and flow in unregulated channels. Where artificial embankments exist, they are often low, discontinuous, and made of heterogeneous materials. In the upper plains, as rivers exit the valleys, they lose lateral confinement, experience slope reduction, and deposit sediments, forming alluvial fans. Here, embankments begin to appear and increase in height and size toward river mouths, where they can reach over 10 m above surrounding ground levels. As rivers approach their mouths, flow velocity decreases, leading to sediment deposition and riverbed aggradation. This process results in “suspended rivers”, where the channel is elevated above the adjacent terrain. In the alluvial plains, rivers are confined by nearly continuous embankments that extend to the coast, transitioning from private to state jurisdiction near urban areas.

In Italy, embankments are classified into five categories according to Royal Decree No. 523/1904, which also defines the responsible management authority. Embankments in Categories I and II are managed by the State and correspond to structures of national and regional (formerly provincial) importance, respectively. As the category number increases, the institutional relevance and scale of the hydraulic structures decreases. The embankments affected by the flood events analyzed in this study fall either under Category II or are unclassified. In the latter case, maintenance and management responsibilities lie with private landowners.

The embankment system of the Po Valley has evolved through repeated interventions, often following major flood events. Embankments have been extended and raised to increase safety margins—for instance, to 0.8 m above peak hydrometric levels observed during the 1801 flood. From 1813 to 1981, the total embankment length increased from 760 km to 953 km, while the length of embankments along Po tributaries grew from 729 km in 1874 to 1257 km in 1981. As a consequence, peak water levels have also increased: at Pontelagoscuro, about 20 km from the Po River mouth, the maximum level rose from 2.19 m in 1801 to 4.28 m in 1951 [26]. This alluvial region is also affected by subsidence, of both natural and anthropogenic origin. The latter is mainly due to the extraction of shallow and medium-deep groundwater for industrial use, as well as methane extraction from deep gas reservoirs. Between the mid-20th century and the late 1970s—when these activities began to decline—cumulative subsidence reached up to 3.5 m in the more affected areas of the plain (e.g., [29,30]).

3. Materials and Methods

The exceptionally intense meteorological events of May 2023 severely tested the resilience of infrastructures across the region, particularly river embankments. Extensive flooding of leveed areas occurred due to numerous breaches and overflow events.

The collection and analysis of data on major breaches and damages were carried out through a coordinated effort involving the Po River Basin Authority (Parma, Italy)—responsible for maintaining the Po River and its tributaries—the University of Bologna (Bologna, Italy), and the Regional Security and Civil Protection Agencies (Bologna, Italy). This collaboration combined knowledge, technical expertise and operational resources to assess the performance and failures of embankment systems during the crisis.

3.1. Methodology for Database Implementation

Given the high number of breaches and the extensive length of damaged riverbanks, data collection was necessarily limited to the most significant cases. The selection of “major” breaches and damages to include in the database was made according to the indications of the technicians of the Regional Agency for Territorial Safety and of the Civil Protection (Bologna, Italy), following these criteria:

• The breached section belongs to a river embankment classified as Category II (see Section 2.3).
• The breach, even if affecting an unclassified embankment, caused significant damage (e.g., flooding of densely populated areas) or occurred near a highly exposed area.
• The presence of peculiar local conditions, such as interference between the embankment and structures (encroachments), like pipes, bulkhead, flood wall, etc.
• Damages that affect large stretches of river embankments, resulting in conditions at the verge of collapse but without the complete formation of a breach, especially in highly exposed areas.

Breaches excluded from the analysis mainly occurred in hilly areas, affecting private and discontinuous embankments, typically constructed solely to protect agricultural fields. Data collection activities included the preliminary design of a standardized format to be used during on-site surveys (Figure 9), referred to as the “Major Breaches and Damages Survey Form”. Information was gathered through multiple steps, listed in chronological order:

• On-site inspection of affected section and, when possible, gathering eyewitness accounts from local residents.
• Meetings with authority technicians to verify the data and fill information gaps.
• Collection and cataloguing of photographic and video material recorded during the flood by citizens, technicians, journalists, etc.
• Retrieval of pre-event information, also through available drone and satellite surveys.
• Import of acquired data into a GIS environment (QGIS, version 3.22.12).
• Compilation of the collected information into the standardized survey form.

3.2. Survey Form of Major Breaches and Damages

The primary purpose of the survey form is to guide the on-site assessment, ensuring consistent and objective data collection regardless of the operator, while emphasizing key information needed for future analyses and understanding the mechanisms behind breach formation and evolution.

Each breach and damage are identified by a code consisting of five fields, providing the following information: type of observed phenomenon (B = Breach; D = Damage), river (first three letters of the river’s name), flood event number that caused the breach (01 = first event; 02 = second event), riverbank where the phenomenon occurred (RG = right hydraulic bank, LF = left hydraulic bank), and progressive breach number (with numbering proceeding from downstream to upstream). This code serves as the title of the monographic form and the reference used in the GIS environment. The structure of the form is divided into three sections schematically represented in Figure 9.

Specifically, Section 1 must report all elements that may have contributed to the breach formation. In particular, it highlights the following:

• The type of material composing the embankment body (if still observable);
• The presence of non-homogeneous layers, such as gravel strata within the embankment, or of interacting structures (manholes, drainage pipes, incorporated fragments of old walls, etc.)
• Extension of the breach and/or damage;
• Identification and description of damages: in case of external erosion, the location and extent of the eroded areas; if the damage is characterized by a crest or body fracture, the width and length of the fracture, as well as the extent of any settled or displaced portion of the embankment (macro or micro-instabilities);
• The possible presence of animal burrows in the area (either visible or reported) and deep rooted plants;
• Evidence of overflow and overtopping, such as bent vegetation and the direction of bending (indicative of the flow direction). It should be noted that overflow may have occurred either, typically, toward the countryside or, exceptionally, toward the river, as observed along stretches of the Lamone, Ronco, and Senio rivers;
• Any particular morphological features of the river section (e.g., meander, narrowing of the cross-section, aggraded riverbed etc.);
• Any other element deemed significant.

In Section 2, the following data are reported:

• Information on the nearest hydrometer(s). The exact locations (in geographic coordinates) are specified, along with the corresponding gage datum. The flood hydrograph responsible for the breach or damages is included, with the critical reference hydrometric levels (green, orange, and red alert thresholds) clearly indicated to allow quick interpretation of the peak water levels reached throughout the event.
• A detailed description of the event chronology leading to the breach or damage formation is provided. Key elements include the date and time of the triggering phenomenon. The timeline of the events can be reconstructed through systematic data collection, including photos, videos, eyewitness accounts from local residents, and real-time reports from journalists and social media channels. Videos are attached to the report (e.g., via links), while photos are included in Section 3. When available, pre-event photos and videos of the river embankment should be collected, as they are particularly valuable for understanding the initial configuration and site conditions prior to the event.
• Photos and details of the in situ collected soil samples. When possible, sampling should be performed ensuring undisturbed conditions through the use of proper sampling tools. Sampling is performed on the intact vertical walls of the breached section at different progressive depths (top, center, bottom) and, if identifiable, the material must be recovered also from the piles of eroded material of the failed section in proximity of the breach.
• Relevant information extracted from technical documentation on past on-site surveys and interventions on the bank structure.

Appendix A shows the application of the Survey Form to one of the main breaches of the database occurred on the Savio river (B_SAV_02_RG_05) in the Regionale-Romagnolo basin.

3.3. Data Organization in a GIS Environment

The identified embankment sections with significant breaches or damages were mapped in a GIS environment using separate shapefiles, each labelled with the corresponding alphanumeric code. The attribute table for each shapefile includes the following:

• River name.
• Location (municipality, province, region).
• Site inspection date.
• Riverbank side (right or left).
• Direct web link to the respective Survey Form.

An example of an exported GIS map is shown in Figure 10A, illustrating a section of the Lamone River at the confluence with the Marzeno river, near Faenza (Ravenna province).

Faenza was by far the most severely affected during the 2023 flood events, with eight breaches located near the urban area. Specifically, breaches from 9 to 11 are located upstream of the city, breaches 8 and 7 are located alongside the historic center, and breaches 5 and 6 are located immediately downstream. These breaches and overflows caused extensive flooding, with water depths reaching up to 8 m in some low-lying districts (Figure 10B). The flooding affected 6000 houses, caused one fatality, and displaced around 2000 people.

4. Results and Discussion

The Italian database has been developed using a structured and consistent methodology specifically aimed at collecting, classifying, and synthesizing data on river embankment breaches, with a focus on the mechanisms, extent, and context of each event. Unlike other existing resources, it adopts a standardized, breach-oriented approach that ensures clarity and uniformity of information. Given the number of breaches documented and the diversity of failure mechanisms involved, this dataset stands as one of the few examples of its kind currently available worldwide—comparable to resources such as the ILPD (International Levee Performance Database, described in [1]) and the USACE National Levee Database [14]—but it is distinguished by the unprecedented consistency and detail in the characterization of breach events, all collected within a single geographic area over two consecutive flood events. The database structure can be easily applied to other regions or used to reorganize existing datasets through a process of data selection, categorization, and harmonization.

4.1. Summary of Breach Occurrence in the Hydraulic Basins

A comprehensive analysis of the database provides a territorial-scale overview of the effects of these two exceptional flood events. Figure 11 summarizes the response of the river embankment system to the May 2023 events, identifying 10 regional rivers that experienced multiple breaches and six rivers with a single breach. It is worth noting the proximity of the rivers that breached at multiple points to several major urban centers. In total, the floods generated 66 major breaches along 16 watercourses: Gaiana, Quaderna, Idice, Santerno, Sillaro, Senio, Lamone, Marzeno, Montone, Ronco, Savio, Rabbi, Lavino, Rio Casalecchio, Uso, and Marano (Figure 11).

The overview of the spatial distribution of river embankment breaches presented in Figure 11 reveals a concentration in the central-southern portion of the region, in proximity of the transition of river courses from higher gradient terrains (i.e., mountainous-hilly zones) to the lowlands reaches. This pattern suggests a possible geomorphological predisposition to failure in these areas where there is an abrupt reduction in flow velocity due to a reduced riverbed slope leading to sediment deposition and aggradation of the riverbed accompanied by higher hydrometric levels. Often these failed segments result the first hydraulic structures specifically built downstream of unclassified or heterogeneous embankments, or river expansion areas. Additionally, a collapse occurring upstream of the embankment structures leads to a reduction in downstream water levels, resulting in less critical conditions that are less likely to cause further failure. The most severely impacted embankment systems are located in the provinces of Bologna, Ravenna, and Forlì-Cesena. In particular, multiple breaches were observed along rivers belonging to the Reno Basin (Gaiana, Quaderna, Santerno, Sillaro, and Senio rivers) and the Regionale-Romagnolo Basin (Lamone, Montone, Ronco, Rabbi, and Savio rivers) (Figure 11). A single breach occurred along five other watercourses, with the breaches along the Marzeno torrent (Regionale-Romagnolo Basin) and the Idice river (Reno Basin) producing devastating impacts due to their proximity to the urban centers of Faenza and Budrio, respectively (Figure 7). Finally, Figure 11 clearly illustrates the relationship between the higher return periods calculated for the hydraulic basins and the greater concentration of breaches. The basins of the Senio, Lamone (including Lamone and Marzeno rivers), and Montone (including Montone and Rabbi rivers), all characterized by return periods exceeding 500 years, recorded 12 breaches, 12 breaches, and 9 breaches, respectively. These were the most heavily impacted basins, and they produced the most devastating effects on the territory.

Table 1. Summary of breaches distribution across hydraulic basins and rivers, categorized by flood event. The number of breaches involving Category II levees is reported in parentheses.

	River/ Torrent	N° of Breaches	N° of Breaches 1st Flood Event	N° of Breaches 2nd Flood Event	N° of Breaches Unknown Event (1st or 2nd)
Reno Basin	Quaderna	4 (4)	3 (3)	1 (1)
	Gaiana	3 (3)	2 (2)	1 (1)
	Idice	1 (1)	0 (0)	1 (1)
	Santerno	6 (5)	0 (0)	6 (5)
	Senio	12 (0)	5 (0)	5 (0)	2 (0)
	Sillaro	6 (3)	4 (2)	2 (1)
	Lavino	1 (1)	0 (0)	1 (1)
Regionale-Romagnolo Basin	Savio	5 (1)	0 (0)	5 (1)
	Rio Casalecchio	1 (0)	0 (0)	1 (0)
	Montone	5 (3)	1 (0)	4 (3)
	Ronco	4 (0)	0 (0)	4 (0)
	Rabbi	4 (0)	0 (0)	4 (0)
	Lamone	11 (8)	1 (1)	10 (7)
	Marzeno	1 (0)	1 (0)	0 (0)
Marecchia-Conca Basin	Uso	1 (0)	0 (0)	1 (0)
	Marano	1 (0)	0 (0)	1 (0)
Total N° of breaches		66 (29)	17 (8)	47 (21)	2 (0)
Total N° of breaches—Reno Basin		33 (15)	16 (6)	17 (9)	2 (0)
Total N° of breaches—Regionale Romagnolo basin		31 (13)	3 (2)	28 (12)	0 (0)
Total N° of breaches—Marecchia-Conca basin		2 (0)	0 (0)	2 (0)	0 (0)

summarizes for each regional basin and for each river/torrent the number of breaches, distinguishing by embankment typology and flood event.

A summary of the breach distribution between the two events is shown in Figure 12A, clearly highlighting the predominance of failures during the second event.

In fact, 70% of the total breaches occurred during this phase, in line with the greater intensity of the recorded meteorological conditions. Indeed, during the second event, 47 major breaches were recorded, affecting 21 Category II embankments. Among these 47, 4 breaches reoccurred at reconstructed sections that had previously collapsed during the first event (these latter cases were excluded from the failure mechanism analyses discussed in the second part of this study). A total of 17 major breaches occurred during the first event, 8 of which involved embankments classified as Category II. Additionally, for two breaches that occurred on unclassified sections, there is still uncertainty regarding their attribution to either the first or second event. Figure 12A also shows that for both events, a larger proportion of breaches occurred on unclassified embankments, which are not subjected to systematic maintenance or management by the competent authorities. The lack of maintenance, the inherent heterogeneity of these embankments, and their typical location in foothill areas make them more vulnerable to failure.

4.2. Preliminary Interpretation of Failure Mechanisms

The analysis of the information collected in the Survey Forms enabled the attribution of the most likely failure mechanism to each breach. This association was performed through a phenomenological analysis, primarily based on on-site evidence.

However, to further deepen the understanding of these mechanisms, quantitative analysis is necessary. Such investigations should include analytical studies, numerical simulations, and experimental testing, supported by field investigations to characterize the physical–mechanical properties of the embankment materials, foundation soils, boundary hydraulic conditions, and structure geometry. It is relevant to note that while such analyses can certainly provide a greater level of detail, they do not necessarily guarantee increased reliability. In fact, many of the vulnerability factors responsible for failure may be modified or obliterated during breach formation. A preliminary classification was carried out by distinguishing between breaches associated with overflow and those unrelated to it, either as a triggering or contributing factor. This step was necessary to challenge the initial—but overly simplistic—interpretation following the events, which attributed all breaches to exceptional hydrometric levels exceeding the design elevation of the river embankments. The results revealed a more complex scenario. Among the 62 breaches analyzed (from an initial 66, excluding 4 reoccurrences in reconstructed sections), overtopping was confirmed in 34 cases, absent in 21, and undetermined in 7 (Figure 12B). Consequently, nearly half of the failures occurred in the absence of overtopping, suggesting that other mechanisms, unrelated to external erosion at the landside slope, played a significant role.

Figure 13A shows the full set of analyzed breaches, grouped according to the hypothesized failure mechanism. Despite the widespread occurrence of overtopping, this process alone can be identified as the principal cause in only 35% of cases. In 26% of the breaches, other mechanisms—such as internal erosion or global instability—were likely responsible, while in the remaining 31%, failure was attributed to combined mechanisms, where overtopping acted as a contributing but not exclusive factor (Figure 13B). For five cases, it was not possible to assign a specific failure mechanism. Among the most significant types of damage observed near the main breaches is the devastation caused by the partial collapse of the embankment slopes and the erosion of the riverside berms, along extended upstream sections of the levees. Both phenomena occurred in rapid succession following the opening of the breaches, having been directly triggered by them.

Their onset can be attributed to two main factors: (i) the intense erosive action of high-velocity flows acting on the river-facing slope and adjacent berm; and (ii) the instability of the riverside slope caused by rapid drawdown following the breach formation and the consequent abrupt fall in water level.

A particularly illustrative case is the damage observed upstream of the Ponte Motta breach on the Idice river, where approximately 3 km of the right bank and 3.3 km of the left bank were affected (Figure 14). A comparison of river embankment cross-sections before and after the flood events indicates that between 500,000 m³ and 600,000 m³ of soil were eroded and transported downstream.

4.3. Statistical Distribution of Breach Widths

Figure 15 presents a statistical comparison of breach widths as a function of the assigned failure mechanism. Distinct patterns emerge in terms of distribution and variability. Figure 15A illustrates the aggregated distribution of all data, including 60 breaches. The overall distribution closely resembles that of the erosion-only group (Figure 15E), suggesting that the inclusion of slope instability does not significantly influence the general statistical behavior of breach widths.

The group comprising breaches caused by internal, external, and mixed erosion mechanisms (Figure 15E, data population n = 47) exhibits a right-skewed distribution with a clear mode and median at 30 m, a mean of 36.9 m, a standard deviation of 23 m, and a coefficient of variation (CV) of 0.63, indicating moderate to high variability. The distribution is influenced by several long breaches (>100 m), which raise the mean. Focusing on internal erosion alone (Figure 15B, n = 15), the data show the most consistent behavior. Breach widths are tightly clustered, with a mean of 30.5 m and a CV of 0.48. In contrast, breaches attributed to external erosion (Figure 15C, n = 22) show the highest variability, with a mean width of 43.5 m, a standard deviation of 30 m, and a CV of 0.68. Despite a median and mode similar to the other groups (30 m), the wide range of values reflects the inherent unpredictability of breaches driven by external erosion.

The lower dispersion in Figure 15B suggests that internal erosion tends to produce more predictable breach sizes. This may be due to the limited number of external variables influencing breach development once initiated, compared to external erosion, which is affected by numerous factors, including levee geometry, rainfall intensity, sediment load, flow turbulence, wind action, and vegetation type. Breaches resulting from combined internal and external erosion (Figure 15D) are associated with the lowest relative variability across the dataset. This category shows a mean width of 32.2 m, a standard deviation of 14 m, and a CV of 0.43, indicating that the interaction between internal and external processes may exert a constraining effect on breach dimensions, leading to more stable outcomes. Finally, Figure 15F presents descriptive statistics for two small subgroups of breaches (each n = 4). The first includes cases involving external erosion combined with slope instability, while the second adds internal erosion combined with slope instability and external erosion. Both subsets yield a mean width of 33.8 m and a CV of 0.56, indicative of moderate variability. Although the small sample size limits generalizability, the results suggest that slope instability, whether isolated or in combination with erosion mechanisms, may not substantially alter breach size distribution.

In summary, while breach widths tend to cluster around 30 m, the governing failure mechanism appears to significantly affect their distribution. External erosion leads to the greatest variability, while internal erosion and combined erosion mechanisms yield more constrained and homogeneous breach sizes. These findings emphasize the importance of developing mechanism-specific estimates of breach width, to improve flood hazard mapping and support more accurate modelling of breach initiation and evolution.

5. Conclusions

This study analyzes the effects of the May 2023 extreme meteorological events on the river embankment systems in the southeastern basins of Emilia-Romagna, with a particular attention to their performance in terms of breaches and damages. The exceptional events generated a unique dataset for investigating the real-scale behavior of such flood protection structures under extreme hydraulic conditions.

The construction of the breach database focused on four key aspects: (a) the selection of criteria used to identify the most relevant breaches to be included in the study, (b) the data collection methodology, (c) the assessment of data quality and representativeness, and (d) the organization of the information within a GIS-based system. The breach selection process was carried out in collaboration with the Regional Agency for Territorial Safety and Civil Protection of Emilia-Romagna, applying criteria based on embankment classification, extent of damage, location within critical areas and interactions with other infrastructures.

Out of more than 100 reported breaches (an exact count remains uncertain due to the presence of private and discontinuous river embankments in foothill areas), 66 breaches of particular relevance were selected. For each of these, a detailed report was compiled, drawing on field inspections, accounts from emergency response personnel and visual documentation from both official and unofficial sources (including newspapers, blogs, citizens and institutional records). By overlapping the estimated return period of maximum daily rainfall on regional hydraulic basins, a correlation between rainfall severity and breach density was identified.

The analysis of the collected data enabled, in most cases, the identification of likely failure mechanisms. Overflow was found to be the direct cause of external erosion and failure in 35% of the breaches, while 26% were attributed to other mechanisms, such as internal erosion and global instability. The remaining 31% were associated with combined mechanisms, in which multiple processes contributed either simultaneously or sequentially. In 8% of cases, the failure mechanism could not be determined due to insufficient information.

The study also identified a potential correlation between collapse mechanisms and related breach width. Specifically, external erosion—when acting alone—tends to produce the highest variability in breach width, while internal erosion and the combination of internal and external processes result in less variable breach dimensions. Nevertheless, across all failure types, the median breach width consistently falls between 30 and 34 m, providing a possible reference value for preliminary flood modelling and risk assessment.

The dataset described in this study could serve as a valuable basis for future in-depth analyses aimed at better assessing the vulnerability of specific embankment sections and informing appropriate intervention strategies. Advancing knowledge in this domain is essential to support more effective flood prevention and climate adaptation policies, ultimately contributing to increased territorial resilience against future flood events. Further research should also deepen the investigation into the relationship between failure mechanisms and river embankment characteristics, with the goal of developing and validating robust methods and models capable of analyzing individual failure processes and simulating the river embankment behavior under critical conditions.