We are grateful to the referee for the careful review of our paper. Please see each reply below addressing the general comments of the referee.

Comment #1: Since limiting the meteorological and atmospheric conditions of a specific day (in which the Taisho eruption occurred) to define the worst-case scenario, does not seem to have been given very reliable results. Nonetheless, the input parameters used for the modelling of the worst-case scenario (analysed within this paper), are too much specific, considering atmospheric conditions that are not common in Japan. Therefore, the scenario modelled, could happen, but with a very low certainty.

Response: We found that our description of the dataset is not good enough, and we will explain it more clearly in the paper. Actually, we did not limit the meteorological conditions in contrast to what the reviewer pointed out. Instead, we simulate ash dispersal of the Taisho eruption with diverse weather conditions for 64 years (1958-2021, 23,736 simulations) to obtain meteorologically-diverse ash deposits in the ground and airborne ash concentration. Given the size of the dataset, for example, it can be used to evaluate conditional ashfall risk in Japan when the Taisho eruption occurs in contemporary settings; estimating the volcanic ashfall risk by analysing both the dispersal and deposition pattern volcanic ashfall over a long period; developing hazard maps for the disaster risk management process; and seeking a better comprehension of the tendency from ashfall distribution from a large eruption over a large region. Following this, we would like to improve the dataset descriptions in the paper, specifically in the introduction section and Sect. 4.

Comment #2: In fact, the validation presented was not satisfactory, and the authors knows it...which was probably due to the fact that the authors pushed the model to a very narrow scenario, forcing the inputs, trying to recreate an historic eruption with sparse data.

Response: We found that we did not explain the validation mechanism in a well-structured manner and we plan to explain it better. For smaller-scale eruption cases, the validation of ash dispersal simulation using the PUFF model for Sakurajima eruptions has been done successfully before (e.g. Tanaka and Iguchi, 2019; Tanaka et al., 2020). However, the validation for the ash dispersal simulation from a large-scale eruption in Sakurajima is yet to be done, and we assume a similar process should work for such a case.

For our validation structure, we assume that if we could simulate the Taisho eruption using complete wind data at the time the eruption took place, we could produce a similar ash distribution for entire Japan. However, it is noted, in the paper, that the Taisho eruption occurred more than a century ago under extraordinary weather conditions. Moreover, we found that complete wind data at the eruption time are not entirely available for us to replicate the eruption. Therefore, to alleviate the issue, we tried to find the dates with similar weather to the time of the eruption, expecting it to have a similar wind pattern. Finally, we checked the simulation result of the selected dates with the available ground reports. Therefore, it is critical for us to highlight the extraordinary weather conditions when the Taisho eruption took place to find dates with similar wind patterns due to the same weather characteristics. In the paper, as we described in the Sect. 5.2, we use weather chart similarity to find such dates. However, we acknowledge that there may be other methods to do this. For example, global climate models may obtain more precise wind conditions during the eruption. Currently, we do not have access to such models, but we will try to incorporate them in the future.

As pointed out in the paper, using the method described there, we can only find a small number of dates that correspond with the weather of the eruption day. Therefore, we only compare the simulation results of the selected dates as a validation mechanism for our dataset. Nonetheless, the validation results of the simulation on selected dates show sufficient accuracy. Therefore, it is estimated that the dataset would also have reasonable accuracy with the ground truth, if available, for different weather conditions.

The validation mechanism we did above does not mean limiting the dataset to only contain specific condition. Following this, we would like to revise the validation structure of the dataset in the paper, specifically in Sect. 5.

Comment #3: The paper is well written, however, the objective of the paper should be reconsidered, along with the scope of the modelling.

Response: We acknowledge the lack of clarity on the objective of this paper that should cover a better dataset description as well as the significance of the dataset.  If allowed by the editor, we would like to add a new subsection for illustrating the usage example of the dataset and the suggestion for the prospective users, specifically in subsection 4.2 within Sect. 4.

References:

We are grateful to the referee for the careful review of our paper. Please see each reply below addressing the general comments of the referee.

Comment #1: Reproduction of ash dispersal from past eruption is an exercise already published elsewhere and for a number of past eruptions. I have to say with better results than for the Taisho eruption. This is probably due to the poor constraints on actual deposits and the broad assumption on meteorological model used for running the simulations.

Response: We recognised that various publications already discuss the reproduction of ash dispersal from past eruptions. However, previous studies on the ash dispersal from a large-scale eruption in Sakurajima volcano did not consider the impact on a broader space and distal locations in Japan. As we explained in Sect. 2, the Taisho eruption is regarded as the typical eruption that a large-scale eruption in Sakurajima volcano would resemble. Based on such a finding, reproducing the Taisho eruption in contemporary weather will provide a framework to prepare responses for the upcoming ashfall disasters to entire Japan. As stated in the paper, we only used one eruption scenario, but we did not limit the simulations to a particular atmospheric condition only. This approach is useful for determining the upper limit value on the ash dispersal as we set the parameter for the latest large-scale eruption of Sakurajima volcano in any given weather scenario (Bonadonna, 2006). We understand that our deterministic approach limits the extent of presenting all possible outcomes of a large-scale eruption. We intend to improve our results with better methodology in future publications.

Comment #2: Modern studies on hazard from ash dispersal use thousand of simulations to produce probabilistic maps of ash deposition, able to capture the meteorological variability. Why to consider only the conditions of a given day? What are the meaning of this map in the light of ash fall hazard? Indeed, the probability to have the chosen condition for a future eruption at Sakurajima are very low. So in which way these simulations may help in drawing hazard maps?

Response: We found that our description of the dataset is not good enough, and we will explain it more clearly in the paper. Actually, we did not consider only the conditions of a given day in contrast to what the reviewer pointed out. Instead, we did the simulation for 64 years (in total, 23,736 simulations) to obtain meteorologically-diverse ash deposits in the ground and airborne ash concentration. We recognised that we did not explain the validation mechanism in a well-structured manner which can be confusing, and we plan to explain it better.

For smaller-scale eruption cases, the validation of ash dispersal simulation using the PUFF model for Sakurajima eruptions has been done successfully before (e.g. Tanaka and Iguchi, 2019; Tanaka et al., 2020). However, the validation for the ash dispersal simulation from a large-scale eruption in Sakurajima is yet to be done, and we assume a similar process should work for such a case. For our validation structure, we assume that if we could simulate the Taisho eruption using complete wind data at the time the eruption took place, we could produce a similar ash distribution to entire Japan. However, we found that complete wind data at the eruption time are not entirely available for us to replicate the eruption. To alleviate the issue, we tried to find the dates with similar weather to the time of the eruption, expecting it to have a similar wind pattern. Finally, we checked the simulation result of the selected dates with the available ground reports. As pointed out in the paper, using the method described there, we can only find a small number of dates that correspond with the weather of the eruption day. Nonetheless, the validation results of the simulation on selected dates show sufficient accuracy.

The figures of the ash distribution map shown in the paper are the results from the selected dates only in the validation section and did not represent the entire dataset. It is understood that the dataset would also have reasonable accuracy with the ground truth, if available, for different weather conditions. Following this, we would like to revise the validation structure of the dataset in the paper, specifically in Sect. 5.

Comment #3: Honestly, I think the manuscript, in its present form, is of poor usefulness for ashfall hazard mitigation over Japan.

Response: We acknowledged the lack of clarity on the objective of this paper that should cover a better dataset description as well as the significance of the dataset.  If allowed by the editor, we would like to add a new subsection for illustrating the usage example of the dataset and the suggestion for the prospective users, specifically in subsection 4.2 within Sect. 4. Given the size of the dataset, for example, it can be used to:

1. Evaluate conditional ashfall risk in Japan (e.g., seasonal and weather-related) when a large-scale eruption, like the Taisho eruption, occurs in contemporary settings.
2. Estimating the volcanic ashfall risk by analysing both the dispersal and deposition pattern of volcanic ashfall for various cases in the last 64 years.
3. Developing ashfall hazard maps for the disaster risk management process.
4. Seeking a better comprehension of the tendency of ashfall distribution from a large eruption over a large region

The dataset is intended to provide information for the local government to prepare the response plans for the ashfall impacts during a large-scale volcanic disaster. The information available in the dataset (ground deposits and airborne concentrations) will help prepare responses against ashfall disasters, such as the creation of conditional ashfall risk maps, early warning system enhancement, planning for ash clean-up operations, closure management on both air and land transportation, and long-term / long-distance evacuation. In addition, other types of users, such as researchers and private sectors, can use the dataset to expand further the research on the mitigation strategies against ashfall hazards.

References:

Bonadonna, C.: Probabilistic modelling of tephra dispersion, in: Statistics in Volcanology, edited by: Mader, H. M., Coles, S. G., Connor, C. B., and Connor, L. J., The Geological Society of London on behalf of The International Association of Volcanology and Chemistry of the Earth’s Interior, 243–259, https://doi.org/10.1144/IAVCEI001.19, 2006.

Tanaka, H. L. and Iguchi, M.: Numerical Simulations of Volcanic Ash Plume Dispersal for Sakura-Jima Using Real-Time Emission Rate Estimation, JDR, 14, 160–172, https://doi.org/10.20965/jdr.2019.p0160, 2019.

Tanaka, H. L., Nakamichi, H., and Iguchi, M.: PUFF Model Prediction of Volcanic Ash Plume Dispersal for Sakurajima Using MP Radar Observation, Atmosphere, 11, 1240, https://doi.org/10.3390/atmos11111240, 2020.

We are grateful to the referee for the careful review of our paper. Please see each reply below addressing the general comments of the referee.

Comment #1: In my opinion, the authors should have used this data set to then develop a probabilistic map of ashfall deposits (and perhaps airborne concentrations at certain times post-eruption) and provided that as a useful hazard map for emergency planners.

Response: We found that we did not put a comprehensive description about the usage and significances of the dataset as pointed out by the referee. If it is allowed by the editor, we would like to add one more subsection within Sect. 4 to explain the usage example of the dataset. As pointed out by the referee, we would like to give some examples of the dataset’s usage in an additional subsection. We can develop probabilistic maps of ashfall deposits and airborne concentrations as recommended by the referee. We will present further dataset analysis in another publication since the paper here focuses more on dataset generation.

Comment #2: I am perplexed by this effort to apply novel techniques to query the meteorological records to find two days with weather that generally match the unusual weather conditions believed to have existed during the Taisho eruption. The Taisho eruption is then modelled with the meteorological conditions of those two days to derive ashfall maps that are compared with the map of ashfall from the Taisho eruption. I find this exercise to be unnecessary and of limited utility, as the PUFF model does not require such validation. Moreover, the map comparisons provided in Figure 15 do not show very good agreement.

Response: We found that our explanation regarding the validation is not well-structured, and we plan to explain it better. We acknowledged the referee’s comment that the PUFF model performance for ash dispersal simulation from a volcanic eruption is well-established.

Our intention in presenting the validation section is to check the model’s performance for replicating the past large-scale eruption; if given the precise wind data on a certain eruption scenario, the model will produce a relevant ash distribution. For smaller-scale eruption cases, the validation of ash dispersal simulation using the PUFF model for Sakurajima eruptions has been done successfully before (e.g. Tanaka and Iguchi, 2019; Tanaka et al., 2020). We think providing a validation framework for the large-scale eruption in Sakurajima is also necessary. For this case, we assume that if we could simulate the Taisho eruption using complete wind data at the time the eruption took place, we could produce a similar ash distribution for entire Japan. However, we found that complete wind data at the eruption time are not entirely available for us to replicate the eruption. To alleviate the issue, we tried to find the dates with similar weather to the time of the eruption, expecting it to have a similar wind pattern. Finally, we checked the simulation result of the selected dates with the available ground reports.

Besides the intention, we understand that the mechanism we utilised is complicated and has limitations, so it may not yield perfect results, both in the wind speed found through the searching process and the ash distribution from the simulation. Therefore, we can follow the reviewer’s recommendation to, perhaps, remove the validation section, but it is also fair for us to show some level of validity for the ash dispersal simulation on a large-scale eruption case like the Taisho eruption. Following this, we would like to revise the validation structure of the dataset in the paper, specifically in Sect. 5.

Comment #3: Sections 5.2 through 5.3 of the paper are an interesting description of applying AI to sort through a complex data set, but anything novel in these sections should probably be published in a separate paper. They have little bearing on calculating the ashfall hazard across Japan.

Response: We acknowledged the lack of clarity on the usage of AI method to find the similar weather to the eruption day. We intend to develop more comprehensive cases of our method in an upcoming publication.

Comment #4: Another general comment is that the entire manuscript should be edited by a native English speaker prior to publication. Portions of the manuscript require extensive editing for proper English.

Response: We recognised the lack of English proficiency in this manuscript. We will use the professional proofreading service to check and enhance the final version of the manuscript after accepted to publication.

References:

Tanaka, H. L. and Iguchi, M.: Numerical Simulations of Volcanic Ash Plume Dispersal for Sakura-Jima Using Real-Time Emission Rate Estimation, JDR, 14, 160–172, https://doi.org/10.20965/jdr.2019.p0160, 2019.

Tanaka, H. L., Nakamichi, H., and Iguchi, M.: PUFF Model Prediction of Volcanic Ash Plume Dispersal for Sakurajima Using MP Radar Observation, Atmosphere, 11, 1240, https://doi.org/10.3390/atmos11111240, 2020.

We put our responses in the original document for the specific and technical comments, directly replying to the comment in the pdf file. If it is difficult to follow, we also provide an alternative response document addressing each comment based on the order of the comments. Please see attached the pdf response files.