| 24 Jun 2026
Flood dynamics driven by multiple levee breaches: two-dimensional hydrodynamic–morphodynamic modelling of the 16–19 May 2023 Senio River event (Italy)
Abstract. The extreme rainfall event that affected the Emilia-Romagna region (Italy) between 16 and 17 May 2023 triggered widespread flooding and multiple levee breaches along the Senio River, leading to extensive inundation of agricultural and peri-urban areas. This study presents a high-resolution two-dimensional (2D) hydrodynamic model based on the parallel GPU-accelerated solver PARFLOOD, coupled with a morphodynamic model to explicitly simulate multiple levee breach formation and evolution during the May 2023 event. The model reconstructs the flood wave using observed hydrometric data and consistent hydraulic boundary conditions. A grid resolution of 1 m enables detailed representation of breach widening, crest lowering, erosion dynamics, and inundation processes. The modelling framework allows quantitative assessment of overflow volumes, flood extent, water depth distribution, and flood arrival times, and is supported by a comprehensive dataset including hydrometric records, high-resolution pre- and post-event topography, flood extent mapping, and eyewitness observations. Results show that morphodynamic feedback strongly controls flood volume partitioning and inundation timing. A substantial portion of the total flood volume is released through breaches; however, return-flow breaches are also observed, whereby part of the water previously spilled into the inundated areas re-enters the river channel. This behaviour leads to complex hydraulic interactions between the river and the surrounding areas, including non-monotonic discharge evolution along the river reach. The analysis highlights the importance of multiple simultaneous levee breaches, a condition that is rarely documented in the literature and challenges traditional flood hazard assessment approaches based on independent breach scenarios. The study shows that a physically based breach approach, coupled with a 2D shallow water model, provides a robust framework for flood reconstruction and offers new insights for residual flood risk assessment in embanked low-gradient river systems.
GENERAL COMMENTS
The paper presents a detailed reconstruction of a sever flooding event occurred in May 2013 in the Senio River (Italy). Given the quantity and quality of data available to assess the event and, above all, the diversity and complexity of the flooding process involved in the flooding event, with multiple (i.e., ten) levee breaches due to both outflow and return-flow forming in a unique flood event, the modelling exercise has the potential to shift the current practice of flood hazard assessment in lowland areas towards more realistic and physically-adherent methodologies that are needed to proper assess relevant processes such as multiple levee failures, which configure as an emergent critical issue.
The paper is clear, generally complete, and well written. I’d like to share some minor comments that could help to further improve the presentation and the discussion of the results.
SPECIFIC POINTS
MINOR POINTS
-l. 16: Please consider "The results show that the morphodynamic feedback involved in levee breach formation strongly controls..." to better introduce the concept of “morphodynamic feedback”.
-l. 34: predefine breach characteristics. From the results and the variability in breach characteristics (shape, size, inner or outer side of erosion, erosion rate different for outflow or return-flow breaches) it is evident that predefined geometries can hardly capture the real evolution of levee breaches. This aspect deserves to be recalled in the discussion or in the conclusions of the paper.
-l. 37-38: simplified approaches to breach formation. It should be acknowledged that "intermediate" approaches exist in which the shaper of the breach is kept fixed (e.g., trapezoidal), but the breach size and depth are evolved in time using simplified erosion models that account for terrain characteristics and hydraulic loading, including possible submerged and reversed flow (e.g., Viero et al., 2013, https://doi.org/10.1016/j.advwatres.2013.05.011)
-l. 42: A sentence briefly describing (i.e., just introducing) the morphodynamic approach to breach formation is missing here when transitioning from high-resolution models to fully coupled hydrodynamic–morphodynamic simulations of levee breach evolution. Previous literature from the same Authors should be added here (e.g., Dazzi et al. 2019; Dazzi & Mignosa 2026, already present in the reference section).
-l. 47, etc.: as a general remark, please consider the use of “overflowing” in place of “overtopping” as, to be specific, overtopping is more suited for wind waves, overflowing for slow-varying cases as river levees.
-Section 2 does not describe only the study area. It could either be moved to the current Sect. 3, as in some sense it reports about "Materials". However, considering that Sect. 3 is already quite long and more focused on methods, consider changing the title of Section 2 from "Study Area" to "Study Area and Flooding Events”. The opening paragraph of Sect. 4, along with Fig. 7, should be placed in Sect.2 (however, I understand the reasons to keep it as closest as possible to the Results section for comparison purposes. So, the choice is up to the Authors).
-l. 64: besides the role of levee on flood containment, many linear infrastructures (e.g., roads, railways, minor channels) in the surrounding, nearly flat areas are expected to affect flood propagation, extent, and intensity, thus requiring suitable inclusion in flood models. Previous literature on that can be referenced here.
-l. 161: delete “though not identical” or explain (if feasible in a stand-alone sentence) which are the differences.
-l. 253: “slope stability checks are not performed at every hydraulic time step but at a larger interval”: How much larger? How is it determined? Taken from previous sensitivity analyses or previous studies?
-l. 341: A reasoning/demonstration about the fact that the shear stresses at the outer slope of leveed and during breach formation largely exceed the critical value for inception of sediment erosion and, hence, model results almost insensitive to variations τc, is reported also in Viero et al. (2013).
-l. 372: At the beginnin of Sect. 4, I would state that the reference is only on the second flooding event (16-19 May 2023), as the first one (1-5 May) was only used to calibrate the model. It has already been said above, but it is useful to recall here.
-Figure 7: Why not use a (slightly) shaded relief to better represent terrain discontinuities and unevenness?
-l. 395: The estimation of flood volumes depends on the timeframe where they are evaluated, so please add this information.
-Figure 8: At flood peak, the discharge difference between S1 and S2 is about 200 m³/s but the green line peaks at almost 100 m³/s. Are there other overflowing occurrences in the reach between S1 and S2?
-Figure 10: The caption is the same as Figure 9. Please update to describe the correct figure. Please also add in the figure the return-flow hydrograph (i.e., the difference between the two plotted hydrographs) in green as in Fig. 8.
-l. 478: “for which a downstream return flow is expected”. In Fig. 7, no return-flow is shown that can be associated with the B2 outflow.
-Figs. 13 and 14: In both these figures please state if the terrain elevation (during and at the end of the event) is the modelled one (the final terrain elevation in Fig. 14 could also have been derived from the post-event LiDAR survey).
-Figure 15: Just a comment. It is absolutely reasonable that the model cannot grasp the soil heterogeneity and thus the irregular profile of the real breach. This probably entails some difference in return-flow discharge, but I see from another plot that this discharge is very small at the end of the simulation (so it is not responsible for the level mismatch at Castel Bolognese in the last 12 hours).
-Figure 16: Please add the boundary of the observed flood extent. In addition, it should be useful to indicate the (or some) significant cross-sections in the figure (if they can be visible according to the zoom level). At least the 3 gauging stations and La Steccaia breach location.
-Caption of Fig. 17 (and others): to reduce the caption length and to improve readability, please consider using such a form: "Maximum simulated water depths in the urban area of Castel Bolognese during the 16–19 May 2023 (colormap), observed flood extent ... (white line), domain used for .... (black dashed line), and locations where... (balck markers)."
-l. 734: I would add a comment to stress that, when the magnitude of inflow hydrographs significantly exceeds the conveyance capacity of a river, the probability associated with the formation of multiple breaches turns out to be non-negligible. A framework to consider the occurrence of multiple breaches on hydraulic hazard assessment has been recently proposed in the literature (Lazzarin et al., 2026, https://doi.org/10.1016/j.jhydrol.2026.136384).
-l. 747: In lowland rivers, my understanding is that unsteady flows lead to a variation of friction slope as a consequence of the different spatial distribution of the flowrate along the river (downstream-decreasing during the rising limb and downward-increasing during the falling limb). This effect is not directly linked to pressure gradients (that, for free-surface smooth flows, is expressed by water level) nor to inertial effects. For the same discharge value, compared to the rising phase, during the falling limb of the flood wave the friction slope is lower because the flow velocity is slower, and this is because the water levels are higher as they are downstream controlled at (downstream) sections where the discharge is higher. This occurs also if inertial effects are negligible. Of course, this sort of backwater effect due to unsteady flow can be ascribed to pressure, but pressure and water level are interchangeable when the pressure is vertically hydrostatic (as is the case here).
-l. 768: Breach opening changes the rating curves because of both discharge reduction and drawdown profiles. Such a conclusion is not new, as it was used, also by some of the same authors (D’Oria et al., 2015, https://doi.org/10.1016/j.advwatres.2015.05.002), to infer outflow hydrographs through breaches by inverse modelling to match the in-river level variation induced by breach opening. I suggest acknowledging this and citing relevant literature.