I have already provided suggestions on this manuscript but they appear as CC instead of RC. This new action is just to fix this.

General comment

The paper provides the use of a hydrologic-hydralic modeling framework, exploiting a fully shallow water equation solutor (implemented in the IBER software) combined with a SCS-CN runoff model in an ungauged basin in Mozambique which was hit by a severe flood in February 2023.

The application is performed by using freely available source of information, i.e. global datasets for DEM, Land cover, CN parameters and precipitation. The basin also includes a large artificial reservoir so that the extent of the flood is affected by spillway regulation whose behavior is analyzed.

The paper is well written and organized and the topic is very relevant with both flood management in areas provided with scarce hydrologic/hydraulic monitoring and also with respect to reservoir management.

As a general comment, I believe that the paper results are interesting and satisfactory with respect to an application developed in a data-scarce environment, nevertheless I would suggest the authors putting maybe less emphasis on the goodness of results achieved and more on the uncertainties related to them, for the reasons explained in following points.

Also, the results of the study case are satisfactory, as just said, but not necessarily easy to straightforwardly address any data-scarce basin in the world. The methodology is, as they say, “reproducible anywhere“ but, I would say, not necessarily with the same rate of success. The success of the application is probably also due to the particular study case which is performed in a (I would say) “enough large” basin (almost 5,500 km^2), with a very low rate of anthropical effect (built area is less than 2%) and a homogeneous topography namely, “a very flat topography”. Moreover, the model validation was possible due to the availability of the Sentinel 1 image of flood extent and by ground observations of maximum water levels in a number of points for the flooded area. To my personal experience, both of these datasets are not always easy to find even in Europe or many other areas of the world. That said, the effort they have provided in finding global databases useful for the application is really noteworthy and also, I would agree that the kind of information they have used can be suggested for best practices of “after event” flood management in any area of the world.

Major comments

#1 The authors provide three scenarios, MS1 (actual management of the reservoir), MS2 (absence of reservoir), MS3 (what if the reservoir was able to retain the entire flow of its upstream basin). The three scenarios provide interesting insights. The comparison between MS1 and MS2 provides that the presence of the dam had a beneficial effect on the flood extent and depth. The third one also testifies that even if the reservoir was larger the AOI would have been flooded by the second tributary (D-PLD subbasin), nevertheless I notice that in this third case the average water level (1.6 m) is significantly lower than in the MS2 (and actual scenario (2.1 m). I agree that the amount of damage could have been not so different, nevertheless a water level less than the average human height could make a significant difference when life is threatened and has to be saved. Hence, I think it could be of interest to know what would have been the flood extent and (average and maximum) depth if, at the initial condition of the event, the reservoir was empty or if it was at a different level below the NPL, see comment #2 below, in order to see if it could be beneficial or not influential at all (as it probably is, due to its small capacity with respect to flood volume).

#2 Figure 7 suggests the need for more information about the operational and the structure of the dam spillway system. In facts, from the figure it appears that in the first four days of observation (6 to 9 February), for water levels up to almost 46.5 m, the daily outflow Qout is zero while, on the 12 and 15 of February for water levels well below 46.5, there is a daily outflow above 500 m^3/s. This observation suggests that the spillways are regulated by some movable device or other hydraulic system that probably were operated (manually or automatically) during the flood event. This is not clearly stated but I believe is necessary for a discussion about reservoir management. The paper does not provide detailed information about the structural and hydraulic operational system for water release. Also there is not a definition of the “Normal Pool Level” (NPL). In particular, it would be of interest to know if there is a minimum level for water release (below NPL) which could be operated by means of such a movable device and, if yes, what is the reservoir volume at that level. These elements could be useful to define a fourth scenario as I have suggested at comment #1

 #3 In section 3.3.3, line 304, a F1 score is mentioned as a combination of HR and FAR, but it is not further defined neither it is used throughout the paper. Also False Negative (FN) cells are mentioned but, if I am not wrong, there is not focus on them in the result sections. I may suggest the use of other indices such as the Critical Succes Index (CSI) and more. Maybe the authors could refine this section by extending the use of these metrics to other indices or explaining the reason why they only focused on HR and FAR.

#4 Figure 9 (right) I would add, besides (or replacing) the regression line, the 1-1 line. The regression line, in facts, provide a satisfactory index of determination but suggests a systematic underestimation of the hydrologic/hydraulic model with respect to observation. By this light I don’t think the regression line provides a correct information. I see that all points but one are almost perfectly centered. Only in one day the daily average discharge is missed (11 February). This could be a lack of the measured precipitation which is almost absent on that day. To my knowledge CHIRPS values of precipitation may have a high rate of uncertainty and also the CN hydrological model used for evaluating infiltration is rather than perfect.

#5 Section 4.1.1. At line 367-369 it is stated that “the positive ME means that the numerical predictions of the maximum water surface elevation have a positive bias with regard to the field estimations, which is coherent with the fact that the water marks identified in the field work represent a minimum threshold reached by the flood”. But at line 280 it is stated “At each point identified, the maximum water depth reached during the flood was estimated”. The authors should clarify whether the points were related to minimum or maximum levels of water. I believe they are maximum levels as it would not make sense to perform a field map of minimum water levels reached by water. I would suggest that the positive bias may be due to a number of different explanations not excluded the hydrological model used for runoff generation. It is well known that the CN method may provide overestimation of both volume and rate of runoff. The upper left portion of Figure 10 provides a shaded area of runoff which is practically all over the D-PLD sub-basin, independently of rainfall intensity which looks quite low in some areas of the sub-basin. Even the rate of infiltration in Figure 9 (left) looks low, even considering the possible underestimation of rainfall which CHIRPS may provide as already said in comment #4.  On the other hand, the high value of the vertical accuracy of the Copernicus DEM (RMSE= 1,7 m) is not good news, considering it is of the same order of magnitude of the average water depth (2.1 m in MS1, 2.9 m in MS2 and 1.6 m in MS3 as from table 6).

#6 Figure 10 (bottom). Here is probably my major concern. It shows the hydrographs computed with Iber at different locations but it seems that the MS1(S3) line is not an output of Iber but rather a linear interpolation of average daily discharges obtained by water levels registered in the reservoir (they are consistent with those shown in Figure 7red line). As a result, the MS1(S2) line here is the sum of an hourly discharge plus a daily discharge interpolated over different values which appears to me as a critical point of the paper. If my considerations are correct I think this point needs re-evaluation by the authors. If we go back to Figure 9 (left) we see that the daily discharge value (3,780 m^3/s) flowing into the reservoir obtained from the IBER output subtends a much larger hourly peak (5,700 m^3/s). That is the same point (the daily average) we find in Figure 7 as the maximum value obtained by IBER as Qin (in light blue). As I noted before we do not know anything about the spillway size and structure and about hydraulic regulation devices but even considering a very high efficiency of such a structure it is hard to believe that the ratio between maximum Qout and maximum Qin is equal to 2700/5700=0.47. In order to sum up the hourly hydrograph of the IBER output from sub-basin D-PLD with the hydrograph of the spillway discharge the hourly distribution of Qout is needed as well. It should be ideally obtained by knowing the geometry of the spillway structure, and of the lake, to route the hourly IBER output Qin of Figure 9 arising from sub-basin U-PLD into the reservoir and then into the spillway in order to obtain an hourly Qout. If such information is not available at least a peak coefficient could be applied to the daily average value shown in Figure 7, or a feasible ratio between hourly values of maximum Qin and maximum Qout should be searched for. Obviously, such consideration also affect results shown as from scenario MS1(S2) (e.g. Figures 13 and 14). It is likewise obvious that, should the authors state that the dam is provided not with a standard spillway system but with a strongly regulated discharge control system, my concern will be solved and with it also the last sentence of the next comment.

#7 Figure 11. In this Figure I see that the northernmost point, ID 6 in Table 3, if am not wrong is the one that provides the highest overestimation (second highest value) in water level h obtained by IBER vs h from field observation: 5.4 m (table 5) vs 2.8 m (table 3) of water depth over the ground level). If I read well Figure 8 this is also the only one placed on a reach affected only by the flood of sub-basin D-PDL. Consider now the significant FAR value (0.37) found in section 4.1.3 and look at Figure 12. I see that a good portion of the False Positive cells affecting FAR are in the same northern reach coming from sub-basin D-PDL. Considering that the discharge coming from the reservoir outflow may be affected by an underestimation error (see #6) both the high FAR value and the overestimation of water depth in point ID 6 may be an effect of the overestimation of runoff arising from the use of the CN infiltration model. Such overestimation may compensate the underestimation of the daily outflow hydrograph from the reservoir in the remaining points.

#8 the rainfall event that generated the flood of February 2023 was particularly severe also by the light of its spatial distribution. In facts, the presence of the highest rainfall intensity in areas close to the reservoir generated a very quick response that did not gave any possibility of operating on the reservoir by releasing water at a discharge compatible with river conveyance with the aim of providing more storage in the reservoir available to mitigate the peak flow. Nevertheless, considering the basin size and travel time of water, a different rainfall distribution may provide this operational time. I would suggest mentioning this possibility, practicable by mean of this hydrologic/hydraulic operational tool, as a discussion item for best practice in reservoir flood management.

Minor comments

Line 85. The CHIRPS acronym is only used in this line and it is not explained. I suggest expanding the acronym and explain in section 3.1.2 the relationship with GPM-IMERG.

Figure 8. What is the shaded area in the background ?

Line 426. The reference, if I am not wrong, should be to figure 14.

The hydrological-hydraulic study presented in this paper attempts to reproduce a pluvial flood across the Umbeluzi catchment in February 2023. The numerical simulations were conducted with the free software Iber+, well known by the primary author because he is one of the original developers. The basin has a drainage area of approximately 5000 km² and a mean slope of about 10%, which led to a peak discharge close to 3000-5000 m³·s^-1 for the studied floods. The study's input data consists of the Copernicus GLO-30 Digital Elevation Model (DEM), 66 satellite-based GPM-IMERG rainfall database pixels, and the curve number (CN) data set GCN250. The spatial resolutions are, respectively, 30 m, 9 km and 250 m. For the validation step, the authors limited the analysis to the outlet region of the catchment using: i) a 10 m resolution Sentinel-1 image, taken on 14 February 2023 at 03:20 UTC, with a discharge of 915 m³s^-1; ii) twenty (post-event) watermarks measured in field works on 20-21 March 2023.

The paper is well presented, the materials and methods are explained briefly, referencing other literature for details, and the results are described concisely. I have no concerns regarding the writing and presentation. However, the size of the basin and the magnitude of the flood could be more exceptional regarding other studies also conducted with Iber+ by other authors not cited in the current version of the paper. The pluvial inundation in the Umbeluzi basin has no particular value because the peak discharge is not high for the catchment size; however, if the authors could show an essential novelty regarding the methodology from a broader scientific perspective, it would deserve publication. Furthermore, the Conclusions are not supported by the Results. As explained below, it is impossible to achieve the Conclusions established in the last section of the paper because of the absence of more accurate input data.

I provide below a series of constructive remarks that can contribute to addressing the weakness of the paper:

1. Introduction. The Authors should cite other software for distributed hydrological simulations based on the two-dimensional Saint-Venant equations and GPU acceleration. In particular, TRITON (Morales-Hernández et al. 2021), SERGHEI-SWE (Caviedes-Voullième et al. 2023) and LISFLOOD-FP (Sharifian et al. 2023). Also, the Authors need to establish the limitations of Iber+, which only allows using one GPU. In contrast, other alternatives allow multi-GPU, precisely, to achieve the required spatial resolution in accurate distributed-hydrological simulations.
2. Introduction. The limitations of the numerical study concerning the use of global data source and the limited amount of data for the validation has to be explicitly explained in the Introduction. Please note that the spatial resolutions you used, i.e., 30 m for DEM, 9 km for rainfall and 250 m for CN, are too coarse for flood hazard mapping using the 2D Saint-Venant equations. In Spain and other European countries, we have made great efforts and spent huge amounts of money to acquire LiDAR data with the accuracy required for accurate flood risk mapping (Díez-Herrero et al. 2009; Sánchez and Lastra 2011; Olcina-Cantos and Díez-Herrero 2021). Both in terms of spatial resolution and elevation errors, among other essential factors. The global data source used by the authors cannot yield accurate flood maps. Otherwise, why are we making so many efforts to accurately implement the EU Directive 2007/60 on the estimation and management of flood risk?
3. Introduction. Please cite other studies using Iber+ and other software for flood risk mapping using GPU and distributed numerical simulations in basins of similar size. For instance, Moral-Erencia et al. (2021) computed and validated flood maps using Iber+ in a catchment of about 2000 km², with a mean slope as steep as for Umbeluzi, using a computational mesh with 20 million cells and sub-metric spatial resolution in some river stretches. Also, note that the satellite-based IMERG rainfall data set (the same one used by the authors) underpredicted the accumulated precipitation by 50% in such a study.
4. Section 3.2 Numerical model. “The size of the mesh elements ranged from 25 m in the main river reaches to 80 m in the hillslopes” (Line 213) and “Considering both models and the whole Umbeluzi catchment, the total modelled surface was 5461 km², and the total number of elements was approximately 2.6 million (Lines 223-224)”. The computational grid is too coarse, even coarser than the DEM. The grid size affects as much as the physical parameters in distributed-hydrological simulations, see Caviedes-Voullième et al. (2012). Subsequently, an additional numerical simulation using between 20 and 40 million cells is required. The grid convergence study is a standard requisite in any CFD simulation (Blocken and Gualtieri, 2012). In my experience, considering that the model is already configured in Iber+, this task is not time-consuming. The authors only need to refine the mesh to achieve the maximum number of cells a single GPU allows.
5. Equations (2)-(3). Why did you neglect the Reynolds stresses even in the main river?
6. Equation (4). Please also evaluate the Critical Success Index (CSI) by Bates and Roo (2000) to compare your value with other studies. The CSI is more common than HR and FAR.
7. Figure 11. “Observed vs. computed maximum water depths at the locations indicated in Figure 8”. The maximum absolute error in the computed water depth values is extremely high concerning the field measurements. For instance: hiber=6 m for hfield= 3.5 m, or hiber = 4 m for hfield=1.9 m. Such errors are too severe for a flood study. It shows clearly that the global data source is inaccurate for detailed flood risk mapping, contrary to the author’s statements in the Conclusion section.
8. Figures 12-14 and their corresponding descriptions: Why did you limit the AOI to the basin outlet? The inundation area is too broad and probably covers the whole floodplain (from a geological perspective). Hence, it is easy to match the observed and simulated flood maps. Please include a map of the DEM slope in such an area to check. Conversely, the D-PLD headwater should be more sensitive and exciting for validation. Indeed, other studies, such as Moral-Erencia et al. (2022), verified the inundation maps in the catchment, not only in the outlet region.
9. Conclusions. The limitations of Iber+ for flood risk mapping in basins of 5000 km² should be clearly stated. Commenting both on the inaccuracy of global data source for DEM and precipitation and also because of the maximum RAM of a single GPU, which limits the total number of cells in the computational grid (and hence, the spatial resolution).

References

Blocken, B.; Gualtieri, C. Ten iterative steps for model development and evaluation applied to Computational Fluid Dynamics for Environmental Fluid Mechanics. Environ. Model. Softw. 2012, 33, 1–22.

Caviedes-Voullième, D.; García-Navarro, P.; Murillo, J. Influence of mesh structure on 2D full shallow water equations and SCS Curve Number simulation of rainfall/runoff events. J. Hydrol. 2012, 448–449, 39–59.

Caviedes-Voullième, D.; Morales-Hernández, M.; Norman, M.R.; Özgen-Xian, I. SERGHEI (SERGHEI-SWE) V1.0: A Performance-Portable High-Performance Parallel-Computing Shallow-Water Solver for Hydrology and Environmental Hydraulics. Geosci. Model Dev. 2023, 16, 977–1008.

Díez-Herrero, A.; Laín-Huerta, L.; Llorente-Isidro, M. A Handbook on Flood Hazard Mapping Methodologies; Number 2 in Geological Hazards/Geotechnics, Geological Survey of Spain; Instituto Geológico y Minero de España: Madrid, Spain, 2009; ISBN 978-84-7840-813-9.

Olcina-Cantos, J.; Díez-Herrero, A. Technical Evolution of Flood Maps Through Spanish Experience in the European Framework. Cartogr. J. 2022021, 1–14.

Morales-Hernández, M.; Sharif, M.B.; Kalyanapu, A.; Ghafoor, S.K.; Dullo, T.; Gangrade, S.; Kao, S.; Norman, M.R.; Evans, K.J. TRITON: A Multi-GPU Open Source 2D Hydrodynamic Flood Model. Environ. Model. Softw. 2021, 141, 105034.

Moral-Erencia, J.; Bohorquez, P.; Jimenez-Ruiz, P.; Pérez-Latorre, F. Flood hazard mapping with distributed hydrological simulations and remote-sensed slackwater sediments in ungauged basins. Water 2021, 13, 3434.

Sánchez F.J., Lastra J. Guía metodológica para el desarrollo del Sistema Nacional de Cartografía de Zonas Inundables. Minist. Medio Ambiente, Medio Rural y Marino, 2011.

Sharifian, M.K.; Kesserwani, G.; Chowdhury, A.A.; Neal, J.; Bates, P. LISFLOOD-FP 8.1: New GPU-Accelerated Solvers for Faster Fluvial/Pluvial Flood Simulations. Geosci. Model Dev. 2023, 16, 2391–2413.