Dear Darren Lumbroso,

Thank you for accepting to review our paper and for the initial comments and discussion. We also believe that the application of consequences models brings huge improvements in risk assessment and emergency planning for tailings dam safety. Within this context, we expect that this work may contribute to advances in practical applications based on the topic related state of the art. We agree and will proofread the paper before the resubmission, adjusting the text considering your initial recommendations.

We agree that there are many other factors related to life loss caused by floods and dam failures that should be included in the general discussion. In our paper, we presented the main categories of factors listed by Jonkman et al. (2008). These categories represent a broader view of the phenomena. The specific factors mentioned in the comments are certainly included in these main categories. For example, factors related to the characteristics of the flood and characteristics of people are encompassed by the estimate of the fatality rate factor; the number of people at risk includes factors related to when the flood occurred; and the effectiveness of the alert system and emergency planning is embedded by the effectiveness of the evacuation. But, indeed, we agree that more discussion about these and other factors would aggregate this description of state of art about life loss estimation in the paper.

We considered Newtonian flow in HEC-RAS because the tailings in the reservoir had a volumetric solid concentration of 12 %, which is considered the flow as aqueous according to O’Brien and Julien (1985). These authors stated that volumetric solid concentration in the waste stream is usually higher than 20 %. Under these considerations, using more robust models such as the ones mentioned, or the more recent HEC-RAS version incorporating non-Newtonian flow should not provide significant improvements to the simulation once the flow behavior is similar to Newtonian flow under these conditions. Furthermore, not enough data is available for more accurate simulations considering rheological parameters. According to O’Brien and Julien (1985), the consideration of flow resistance, like Manning adjustment technics proposed by Travis et al. (2012), would be necessary for higher volumetric solid concentration (>20%) which is not the case in this study. However, in the attempt of calibrating the flood map, some sensitivity tests were performed for Manning roughness in order to better represent the real event. This information may be included in the final version of the article.

For representing the evacuation that occurred in Miraí town during the dam failure event, we consulted local studies that qualitatively described what happened. We agree that the organization of the evacuation could be better understood with a better explanation and insertion of a table summarizing all aspects of the alert and evacuation processes, as done by Bilali et al. (2021, 2022). Therefore, for a deeper comprehension of this important aspect of the accident, we consider producing and including this table in the final article.

All the representation of alert and evacuation (warning issuance delay, warning diffusion delay, mobilization time) was based on Sorensen and Mileti's (2015) works. The authors analyzed many cases, not only about floods but mainly about chemical and fire accidents. They proposed many defaults according to the characteristics of the event and the population. Both evacuation and life loss rates used within LifeSim can be modified by the users. So, it is possible to adapt the process of alert and evacuation to incorporate specific characteristics for a specific case, and it is possible to change the life loss rates by incorporating the impact of the debris and other characteristics of tailings in the contact human-fluid. We will insert a brief discussion about this in the paper.

By “a pioneer” in estimating economic damage and loss of life for tailings dams, we meant this work is the first analyzing jointly these types of consequences in tailings dams failures. Some studies worked with economic damages valuation for water reservoirs dams (Lee and Noh (2003), as you remember) and tailings dams (Veizaga et al., 2017), and others considered life loss estimation for water dams (Bilali et al., 2021 and 2022; and Kalinina et al., 2021). Lumbroso and Davison (2021) were the first study to evaluate loss of life for tailings dam failure. However, this study is the first one to estimate and validate both consequences in the same study related to a tailings dam failure.

Finally, indeed the study had many hydrology and hydraulics uncertainties, including the coarse DEM data available, the physic breach register lacks and others. Some of them were reduced by improvements (e.g., the DEM was refined using terrestrial topographic survey data). Furthermore, we evaluated the impacts of these uncertainties on the impacted population estimates, using two different vulnerability scenarios designed according to the observed and simulated extent of the flood. Considering these two scenarios, we noticed that the life loss estimates were not significantly impacted by these uncertainties. Nevertheless, we will improve this explanation in the paper in order to give more comprehension related to the limits of the performed simulations.

Dear Lukas Riedel,

Thank you for accepting to comment on our paper. Indeed, more robust studies about tailing dam failures with the application of consequences models and others are important and bring huge improvements in risk assessment and emergency planning for this type of structure.

We agree that uncertainty analysis in flood modeling is extremely necessary for the flood consequence model. By the way, this is a theme that we are developing in our flood research group at the Federal University of Minas Gerais. Specifically in this study, we tried to represent the real event that occurred in Mirai, in 2007. Our aim was to verify the application of consequence models by comparing real and estimated outcomes. It was possible to collect the real event data mainly based on local technical reports and information collected from local authorities and post-event local studies. The breach characteristics, arrival time at the city and the extension of the flood are examples of this data. Even though the flood extent uncertainty was not the main focus of the study, considering its relevance to the whole process of evaluation, this uncertainty was partially incorporated into the evaluation by considering different scenarios. We evaluated the impacts of the flood model uncertainties on the impacted population estimates using two different vulnerability scenarios designed according to the observed and simulated extent of the flood. It consisted in compensating the differences between simulated and real event flood extent in terms of the number of people exposed. We spatially increased and reduced the amount of population exposed in the buildings nearby areas where differences were observed between simulations and the real event. Considering these two scenarios, we noticed that the life loss estimates were not significantly impacted by these uncertainties. Nevertheless, we will improve this explanation in the paper to give more comprehension related to the limits of the performed simulations.

Concerning the software version used, we first applied the more recent LifeSim 2.0, version released in August 2021. We identified some bugs in the second version of the software which compromised the evacuation module analysis done in this article. LifeSim 2.0 does not run when it is considered a fraction of the population evacuate on foot. Moreover, when we use other parametrization in the alert and evacuation process instead of the defaults present in the model, LifeSim 2.0 also does not run. These bugs were reported to the Risk Management Center of the Hydrologic Engineering Center. We received an answer from the RMC team reporting to us that they will fix it for the next version. More information about the capabilities of the next version can be explored at this link: https://publibrary.planusace.us/#/document/6477d9a0-3e5e-4621-c7d4-de95a4dcc631. Therefore, we opted to use the LifeSim 1.0 version. In our paper, we highlight the main differences between these two versions, but we will reinforce this choice in the final version of the article. Furthermore, it is expected that no significant changes would be observed in the results of the simulations if the more recent version could be achieved without these bugs, once the differences between the versions do not affect the original method of loss evaluation. In general, the method of both versions is the same following the work and data used by Aboelata and Bowles (2005). The differences between them are some considerations about uncertainties sampling; validation efforts; availability of other features like agricultural losses, indirect income and job losses; and another classification (based on the same data) of limits for building submergence and fatality rates.

The O’Brien and Julien (1985) study is available at this link: https://www.engr.colostate.edu/~pierre/ce_old/Projects/Paperspdf/O%27Brien-Julien%20UtahPDF.pdf. The authors defined classes of flow type by several experiments performed at Colorado State University on mud flow samples. We will add more discussion about this study and others of the same authors to emphasize our option of considering a Newtonian flow for the propagation of dam failure flood. Despite this consideration, we achieved results quite similar related to the flooded area, taking into account that uncertainty is also present in local topography data.

The main characteristics of the real flood event were obtained from Rocha (2015), available at https://smarh.eng.ufmg.br/diss_defesas_detalhes.php?aluno=1135. The author brought all available details concerning the hydrodynamic flood wave propagation. This information was crucial for parameterizing our flood model and for the vulnerability and exposure analysis. We will make it clearer in the text and, as you recommend, we provide the link to this work.

For the preprint, we decided to use the option of code and data available on request. Even though, the assets, including the codes, the models, etc.. Within the final version of the paper, we will make it accessible in a public repository as suggested.

Furthermore, for resubmission, we will consider and analyze all technical corrections and suggestions. Thank you a lot for proofreading the paper and for noticing some mistakes and possible improvements in text writing.