Statistical Insights on the Eruptive Activity at Stromboli Volcano (Italy) Recorded from 1879 to 2023

Round 1

Reviewer 1 Report

In this paper, the authors analyse and list the type and date of known events (primary eruptions and secondary phenomena) occurred at Stromboli volcano. They evaluate the quality of the dataset, and perform a statistical analysis to analyse the shape of the distributions and quantify probability of eruption after certain (15 days, month, ..3 years) timespan. The topic is interesting to the community and the dataset shared by the authors is potentially very important, if data are better described in the paper.

I have two major criticisms to this work, which I think should be addressed before publication.

TYPES OF EVENTS

The authors introduce a more detailed categorization of explosive events (Major, Paroxysmal and uncertain explosions), categories of effusive events such as lava flows of small volume, intra crater flows, and flank eruptions. A more detailed characterization requires strict criteria to be reliable, especially when applied to such long term observations.

Lava flow are classified based either on duration or volume. Why is that?

How are U events classified with respect to M? Please specify also commenting on all existing classifications proposed for this volcanic activity.

How did you classify newly identified older, poorly described events ? (table 1a)

Finally, collapse events such as PDC, tsunami, crater failure and landslides are not used for any statistical analysis, they should be cut from this work as of no use in this research.

I welcome introducing effusive events, but I strongly suggest the authors to look for interdependency (or complete lack of ) with explosive activity, which is a point of primary relevance for hazard assessment. Specific statistical test can be done to look for parameters interdependency.

 

 

STATISTICAL ANALYSIS

The major issue of the non homogenous quality of the dataset is shown and acknowledge by the authors in many ways (fig. 3); This is also reflected by the cumulative curves of fig, 4, whose changes of slopes are of unclear origin and thus not explainable.

Statistical analyses are done on the category of ‘explosive events’. My worry is that this could be misleading as in certain periods (see Bevilacqua et al.) smaller events (major explosions, and more likely, U explosions were not recorded. As a consequence, also the total number of events is biased. Same foreffusive events. Do the authors think that small lava flows were recorded efficiently before 1985?

This is not a secondary point because it would be the main factor controlling the heavily tailed distribution shown in fig. 6.

It makes poor sense to me to look as the distribution of inter event times in between 1879 and 2023 as is it is, having said that at some time the activity was not recorded as efficiently as in the last 35 years. I suggest showing only empirical curves for the 1879-2023 period.

This hypothesis could be confirmed by stationarity and self independence tests on the distribution itself. If the time series is no stationary, there is no point in considering the entire time span, but it is necessary to study stationary periods independently.  This is a critical point for results of chapters 3.1 and 3.2. If I remember correctly, previous studies only accounted for paroxysmal events only because of unreliability of the M data.

 

As for the tested models, I have some comments on the Power law distribution: it is scale invariant, so mathematically, it has no typical scale or size, and the calculation on mean values is basically mathematically meaningless: a self scaling function has no mean. So what do we get from that? Probability scales always in the same way, no matter the point of the distribution we are observing. It is used for assessing hazard of many natural events including earthquakes and floods, in terms of expected magnitude of the most likely event in a certain dimension (space, time). Generally, we think that it reflects the internal hierarchy of the physical processes controlling the event dynamics. The authors apply power law equation to model the tail of the distribution, for which the probability is scale invariant. Is it useful? Why ? What is the physical reasoning behind this approach? Usually power law are used to model extreme events, but with many caveats (Geist and Parsons, 2014, Nat Haz, 72)

Honestly, this is the first time I have seen it applied for probability of occurrence of events (i.e. in survival analysis). Usually, we use parametric models which either correspond to proportional hazard (Cox) or accellerated failure models, such as the Loglogistic, Weibull that the authors are testing here.

From fig. 7,8,9, 10 (and as declared by the auhtors) It is clear that power law equations could fit only a part of the distributions, whereas the other tested models better fit the entire range.

Negative log likelihood values of different models cannot directly be applied like it is has been suggested in the paper. Log likelihood depends on the number of variables, it can be applied only when comparing nested models. In the case of the data presented here, it is necessary to use other information criteria, like AIC or BIC. Based on the figure in this paper, I am expecting that this approach will lead to different conclusions. Finally, I believe it is not fully correct to compare the fitting quality of curves to different portions of the empirical dataset. The fitting ‘quality’ should be compared to the same dataset. Generally, from the results, it seems that all the AFT models tested are more or less equivalent in terms of quality of fit.

 

Specific observations:

 

Table 1b add median values, which are more indicative with respect to mean when the distribution is asymmetric

 

The final sentence of the authors (I have to copy it here because of lack of line numbering in the text)

‘However, the most important consequence of this result is not limited to the model accuracy, but rather to the fact that the power-law distribution is characterized by the so-called scale invariance. The lack of a characteristic scale also implies the lack of a characteristic frequency of events. Consequently, if in a certain time interval, a frequency of occurrence is observed that is lower or higher than the average, this should be considered an intrinsic behaviour of the volcanic activity.'

is circular. If there is no characteristc (average) frequency, what is the significance of its fluctuations? do you mean that if the activity changes time distribution it means that the volcano is changing its state?

 

the abstract presents too many introductory data which should be cut. It lacks key findings of the research, which should be listed.

 

 

 

 

 

therea are several typos which should be corrected -

some small corrections:

 

Introduction

Change ‘involving fallout’ to: Involving tephra fallout

Change ‘giving rise to eruptive columns more than 1 km high above the craters’ to ‘forming columns rising more than 1 km above the craters’

 

Author Response

In this paper, the authors analyse and list the type and date of known events (primary eruptions and secondary phenomena) occurred at Stromboli volcano. They evaluate the quality of the dataset, and perform a statistical analysis to analyse the shape of the distributions and quantify probability of eruption after certain (15 days, month, ..3 years) timespan. The topic is interesting to the community and the dataset shared by the authors is potentially very important, if data are better described in the paper.

 

We would like to thank the reviewer for their work and assessment. We have now significantly improved the description of data used in this paper.

 

I have two major criticisms to this work, which I think should be addressed before publication.

 

TYPES OF EVENTS

 

The authors introduce a more detailed categorization of explosive events (Major, Paroxysmal and uncertain explosions), categories of effusive events such as lava flows of small volume, intra crater flows, and flank eruptions. A more detailed characterization requires strict criteria to be reliable, especially when applied to such long term observations. Lava flows are classified based either on duration or volume. Why is that?

 

We would like to clarify that this paper doesn't propose a new classification scheme for the eruptive activity of Stromboli. We have tried to do so in several previous papers (Calvari et al., 2021; Corradino et al., 2021; Giudicepietro et al., 2022), but the proposed classifications are difficult to apply because they involve too many parameters, and this is the reason why a general consensus has not yet been reached. We have summarized in the Introduction the main classification categories used at Stromboli, using now extensive references. In compiling our catalogue for the explosive activity of Stromboli, we have applied the three categories used by Bevilacqua et al. (2020a, 2020b) to avoid confusion. We have changed the classification of very few events when compared to the catalogue of Bevilacqua et al (2020a) only when we had additional information, essentially derived from the INGV-OE monitoring systems and reports, as now explained in the text and clearly reported in the last column of our catalogue (Table SM1).

 

When dealing with effusive activity, we realized that we lack sufficient information to distinguish them. The most common data on lava flows is their duration, which can be considered as a reliable proxy for erupted volume. This is true for all cases, but not for the small lava flows spreading within the crater. These tiny lava flows can last several days, but result in extremely small lava flows (tens of meters long, directly observed) due to the very low effusion rate (much less than 0.1 m³/s). Thus, the criterion that we can apply is merely qualitative, and refers mainly to the duration of the lava flows for flank eruptions and overflows, where the duration is a good proxy for erupted volume. Conversely, we have decided to consider an additional class for the tiny flows spreading within the crater, taking into account their small volume rather than the duration. We have better clarified this aspect in the new text.

 

How are U events classified with respect to M? Please specify also commenting on all existing classifications proposed for this volcanic activity.

 

U (uncertain) explosive events are those in between major explosions and persistent explosive activity. We have considered U those that are mentioned in the activity reports as "strong explosions" or "explosions more powerful than normal", but for which we don't have enough data to be classified as major explosions. We have better clarified this issue in the new version of the manuscript.

 

How did you classify newly identified older, poorly described events? (table 1a)

 

Thanks a lot for having pointed out this aspect. We have improved the description of the classification of events in this paper. We have considered them as major explosions, uncertain or paroxysms if they are classified as such in the INGV activity reports or in the references cited in Table SM1, or from the analysis of the videos recorded by the INGV-OE monitoring network. In addition, the last column of Table SM1 contains all the references that we have used to build up our catalogue.

 

Finally, collapse events such as PDC, tsunami, crater failure and landslides are not used for any statistical analysis, they should be cut from this work as of no use in this research.

 

We have dedicated several months to compiling the catalogue presented in this paper, and we have recognised only at the end of our analysis that we cannot apply any statistics to some of the collected data because they are too few to be significant. However, in Figure 4c and 4d we have shown that these events are becoming more frequent, which is an important conclusion for hazard purposes, and we have reported this important observation also in the abstract. As this database is unique, these data may be completed and used in the future in case more of such events occur, thus we prefer to keep the catalogue complete as it is now.

 

I welcome introducing effusive events, but I strongly suggest the authors to look for interdependency (or complete lack of ) with explosive activity, which is a point of primary relevance for hazard assessment. Specific statistical test can be done to look for parameters interdependency.

 

We thank the reviewer for this very important suggestion. Testing for statistical interdependency is not an easy task and some methods have been proposed in literature. Due to the short time available for making the revision, we have simply computed the correlation coefficient between the two time-series daily sampled, which has 0 if no event occurred and 1 on the contrary. The two vectors, for the explosive and effusive activity, respectively, have equal length of 52658, i.e. the number of days from 04 Feb 1879 and 07 Apr 2023, which are the first and last dates in the considered dataset.

The correlation coefficient between the two vectors is 0.1393, which indicates a rather low degree of interdependency. However, since we are not sure that we can definitely exclude an interdependence between explosive and effusive activity based on this simple test, we aim to delve deeper into this important aspect in our future analyses.

 

STATISTICAL ANALYSIS

 

The major issue of the non homogenous quality of the dataset is shown and acknowledge by the authors in many ways (fig. 3); This is also reflected by the cumulative curves of fig, 4, whose changes of slopes are of unclear origin and thus not explainable.

 

Statistical analyses are done on the category of ‘explosive events’. My worry is that this could be misleading as in certain periods (see Bevilacqua et al.) smaller events (major explosions, and more likely, U explosions were not recorded. As a consequence, also the total number of events is biased. Same for effusive events. Do the authors think that small lava flows were recorded efficiently before 1985?

 

No, you are right, and we have several times pointed out this aspect in our analysis. And in fact these lava flows had to be removed from the statistical analysis (see for example Table 5). But we had to try at first in order to verify.

 

This is not a secondary point because it would be the main factor controlling the heavily tailed distribution shown in fig. 6. It makes poor sense to me to look as the distribution of inter event times in between 1879 and 2023 as is it is, having said that at some time the activity was not recorded as efficiently as in the last 35 years. I suggest showing only empirical curves for the 1879-2023 period.

 

Despite the well-founded suspicions of sampling shortcomings for the entire period (1879-2023), especially for minor events (ME and U), we believe that having estimated the CCDF probabilities for this time interval is not entirely without use. This study helps to highlight that if it is assumed that the behaviour of the volcano has not undergone major changes from 1879 to today, the differences between the models must be attributed to sampling deficiencies in the entire period (1879-2023) compared to the most recent period (1985-2023).

 

This hypothesis could be confirmed by stationarity and self independence tests on the distribution itself. If the time series is no stationary, there is no point in considering the entire time span, but it is necessary to study stationary periods independently. This is a critical point for results of chapters 3.1 and 3.2. If I remember correctly, previous studies only accounted for paroxysmal events only because of unreliability of the M data.

 

This is not correct. Bevilacqua et al 2020 considered both paroxysms and major explosions together, because the first class comprised too few events.

 

 

As for the tested models, I have some comments on the Power law distribution: it is scale invariant, so mathematically, it has no typical scale or size, and the calculation on mean values is basically mathematically meaningless: a self scaling function has no mean. So what do we get from that? Probability scales always in the same way, no matter the point of the distribution we are observing. It is used for assessing hazard of many natural events including earthquakes and floods, in terms of expected magnitude of the most likely event in a certain dimension (space, time). Generally, we think that it reflects the internal hierarchy of the physical processes controlling the event dynamics. The authors apply power law equation to model the tail of the distribution, for which the probability is scale invariant. Is it useful? Why? What is the physical reasoning behind this approach? Usually power law are used to model extreme events, but with many caveats (Geist and Parsons, 2014, Nat Haz, 72)

 

We have clarified the usefulness of power-law distributions to model our dataset in the Methods section: “The use of the power law distribution as alternative to others probability distribution models has been recognized valid because various natural phenomena, including the field of geosciences, seem to follow this type of distribution, even if a rigorous statistical analysis is particularly complex. The difficulties arise because the power-law is valid for a portion of the dataset, which normally escapes the smaller events, due to the difficulty of detecting them, and larger ones, due to the finite size of the physical systems. Bak (1996) has first discussed the ubiquity of power-law distributions in natural systems. To refer the vast literature concerning this kind of distribution is a hard task and it is beyond the scope of this paper. Limited to the application of power-law models in geoscience it is possible to refer the review paper by (Corral and Gonzales, 2019).”

 

Honestly, this is the first time I have seen it applied for probability of occurrence of events (i.e. in survival analysis). Usually, we use parametric models which either correspond to proportional hazard (Cox) or accellerated failure models, such as the Loglogistic, Weibull that the authors are testing here.

 

From fig. 7,8,9, 10 (and as declared by the auhtors) It is clear that power law equations could fit only a part of the distributions, whereas the other tested models better fit the entire range.

 

Yes, correct. In the paper, it is clearly stated that the power law model can only fit part of the distribution while the alternative distribution can fit the whole dataset. But it is also clearly stated that the power law fits better than the alternative the tail of the distribution, which usually comprises the most energetic events. We have improved the text to better relay this message.

 

Negative log likelihood values of different models cannot directly be applied like it is has been suggested in the paper. Log likelihood depends on the number of variables, it can be applied only when comparing nested models. In the case of the data presented here, it is necessary to use other information criteria, like AIC or BIC. Based on the figure in this paper, I am expecting that this approach will lead to different conclusions. Finally, I believe it is not fully correct to compare the fitting quality of curves to different portions of the empirical dataset. The fitting ‘quality’ should be compared to the same dataset. Generally, from the results, it seems that all the AFT models tested are more or less equivalent in terms of quality of fit.

 

Following the reviewer's suggestion, we have adopted the AIC index as goodness of fit indices. However, the AIC has confirmed that the power-law models fit better than the alternatives the tail of the dataset.

 

 

Specific observations:

 

 

 

Table 1b add median values, which are more indicative with respect to mean when the distribution is asymmetric

 

Yes, thank you. We have now reported the median instead of the mean.

 

The final sentence of the authors (I have to copy it here because of lack of line numbering in the text)

 

‘However, the most important consequence of this result is not limited to the model accuracy, but rather to the fact that the power-law distribution is characterized by the so-called scale invariance. The lack of a characteristic scale also implies the lack of a characteristic frequency of events. Consequently, if in a certain time interval, a frequency of occurrence is observed that is lower or higher than the average, this should be considered an intrinsic behaviour of the volcanic activity.'

 

is circular. If there is no characteristc (average) frequency, what is the significance of its fluctuations? do you mean that if the activity changes time distribution it means that the volcano is changing its state?

 

Yes, we agree with the reviewer. We have rewritten the sentence to clarify the main implication of the power law distribution and lack of characteristic frequency in the considered problem. Paraphrasing the reviewer's suggestion, we have rewritten the sentence as follows:

 

“However, the most important consequence of this result is not limited to the model accuracy, but rather to the fact that the power-law distribution is characterized by the so-called scale invariance. The lack of a characteristic scale also implies the lack of a characteristic frequency of events. In our problem, this implies that the average inter-event time is not relevant for the hazard assessment, similar to what happen to many natural events including earthquakes and floods in terms of expected magnitude of the most likely event, both in time and space”.

 

 

the abstract presents too many introductory data which should be cut. It lacks key findings of the research, which should be listed.

 

We consider it important to introduce the problem before giving the results, that are summarized in the last sentence of the abstract: "Our results confirm the recent trend of a significant increase of major explosions, small lava flows and summit crater collapses at the volcano, and might help monitoring research institutions and stakeholders to evaluate volcanic hazard from eruptive activity at this and possibly other open-vent active basaltic volcanoes."

Comments on the Quality of English Language

 

therea are several typos which should be corrected -some small corrections:

 

We have asked our institutional English mother-tongue reviewer to correct the English style before re-submitting these corrections.

 

Introduction

 

Change ‘involving fallout’ to: Involving tephra fallout

 

Done. Thank you

 

Change ‘giving rise to eruptive columns more than 1 km high above the craters’ to ‘forming columns rising more than 1 km above the craters’

 

Done.

Reviewer 2 Report

General observations

1.       One may wonder what is the reason the manuscript was proposed to the journal Remote Sensing and not to a volcanology-specialized journal?  This is not criticism!

2.       The manuscript follows previous works of the same authors and other authors, correctly mentioned and cited, whose published data are incorporated in a much larger (numerically and phenomenologically) data set completed and updated to the end of June 2023 (!) by the authors themselves allowing a statistically more saturated, hence more precise and relevant interpretation of all available data (old and new) related to the eruptive history of Stromboli volcano from 1879 onwards. Large data set analysis and processing benefit from adequate statistical tools used by the authors.  As so, the paper is a welcome and valuable contribution to the knowledge of eruptive activity of the volcano, its subtle time-dependent variation allowing better hazard assessments to be obtained in the future.

3.       English language expression of the text is at a pretty good level (as far as a non-native English-speaking reviewer can appreciate it), but same language editing would be welcome.

4.       My only major critical observation is related to manuscript organization in chapters and subchapters which I found a bit illogical hence it could be improved. For example:

-          Chapter 2 Data and methods:  Subchapters 2.1 Data and 2.2 Methods should be presented in reverse order, i.e., Methods first, then Data

-          Both subchapters 2.1. and 2.2 contain data, including diagram representation of data (e.g., Fig. 6 in Methods)

-          Data interpretation is introduced in the subchapter (2.1.) of data presentation (e.g., following Table 1b in Page 5): “The main cause of the non-uniform distribution of the events at Stromboli is …”. There are more such cases of data interpretation in 2.1. Data

Therefore, I kindly suggest the re-organization of the chapters/subchapters according to the following scheme (or other similar, in which methods, data and interpretation of data are clearly separated):

1.       Introduction

2.       Methods (including data acquisition sources and procedures)

3.       Results

3.1. Data (processing and presentation in Tables and diagrams, including Numerical results)

3.2. Interpretation

3.2.1 Statistical models for the 1879-2023 data set

                        3.2.2 Statistical models for the 1985-2023 dataset

                        3.2.3 Statistical interpretation of results

4. Discussion and Concluding Remarks

 

5. Since my expertise does not cover the mathematical expression of statistical data processing, this methodological subject is not addressed at all in this review.

 

Specific observations (the text parts to which they refer to are highlighted with yellow in the pdf file)

1.    Introduction

-          “…increasing more than ten times during the summer season” – please add the reason for that (e.g., visitors)

-          Fig. 1 – please mark with a well-visible symbol on the image in b) the actual location of Semaforo Labronzo

-          “… mild Strombolian explosions that occur at a frequency of one every few minutes” – as I know the frequency of this kind of eruptions ranges from a few minutes to a few tens of minutes (witnessed by myself when visited the volcano in 2017)

-          “… this latest event of 19 July 2020 being at the boundary between ME and P…” – how that “boundary” was - or can be - defined (by numbers? else?)

-          What “magma fingering” means? Intrusion? Readers interested in Remote sensing are not familiar with this expression for sure. So, please explain it in a few words in brackets where first used.

2. Data and methods

-          “The details of the monitoring cameras are deeply illustrated by…”. What “deeply illustrated” means? How the details of the monitoring cameras can be “deeply illustrated”?

-          “40 are uncertain major explosions (U), which means that they lie at the boundary between the persistent explosive activity and major explosions”. The same issue as above: how the “boundary” can be defined?

-          Tables should not be divided in two separate pages (the case of Table 1b divided in pages 4 and 5)

3. Numerical results

-          “… we consider the reduced and more recent portion of the catalogue…” – better shortened instead of reduced. If replaced, please do that in the whole text.

-          “… pointed out to an increasing number of …” – more correct: “… pointed out an increasing number of …”

-          “modifications of the shallow conduit” – “plumbing system” is a more appropriate term

3.1 Statistical models for the 1879-2023 data set

-          Tables 3a (explosive events) and 3b (effusive events) can be joined as a single Table 3. If so, references in the text should be adapted to this change; the same for Tables 4, a, b, c, d, and for Figs 5a, b, c.

-          “… there is a slight advantage…” – what “advantage“ means in this context? Unclear.

-          “focus” instead of “concentrate”

-          “… the graphic behaviour of the corresponding power-law models are shown … 1) behavior is correct grammatically; 2) is “behavior” the adequate word here? “graphic behavior” sounds strange to me; 3) “is” instead of “are”

-          “The blue star symbol in the abscissa…” in caption of Fig. 8: the blue star in the figure itself is too small and difficult to be seen. Can you make it bigger?

-          “…are reported in Figure 10a to 10c, for explosive, F effusive and FF+FFF effusive, respectively.” – the word “events” has to be inserted between “effusive” and “,respectively”

3.2   Statistical models for the 1985-2023 dataset

-          “… the results are summarized in Figure 11a to 11c and in Table 6a to 6c…” – commonly, Tables are placed first and the figures showing the data in a graphical form, are placed after them. This observation is valid for the whole manuscript.

-          Again, since the column heads are exactly the same, Tables 6 a, b, c and 7a, b, c can be combined as Table 6 and Table 7, respectively, having three rows each.

3.3   Statistical interpretation of results

-          “… as has also been experimentally demonstrated in this paper.” – please indicate where exactly in the paper that was demonstrated (intra-article reference)

4.    Discussion and Conclusive Remarks

-          “concluding”, instead of “conclusive”, sounds better in the chapter’s title

-          “… we present a new catalogue of events …” –Table SM1 displaying this catalogue has to be referred to

-          “… their occurrence is too low …” – do you mean the frequency of their occurrence is too low?

-          “ … the sizes and capacities of the plumbing systems of the two volcanoes are

-          very diverse …)” – I would suggest: “contrastingly different”, or just “different” instead of “very diverse”

-          “… a contribution to the risk assessment of both explosive and effusive events.” – Correct is “hazard assessment”, because “risk” is related to vulnerability, a subject not addressed in the manuscript

-          “Here, we propose and speculate …” - sounds a bit strange to me, but it is up to you to change

-          “In addition, we have proposed to the well-known heavy tailed distributions ofexplosive activity, already taken into account by previous authors (Bevilacqua et al 2020b) for modelling the inter-event time for the major explosions and paroxysms recorded at Stromboli, the use of a power-law distribution.” – A quite complicated sentence difficult to follow and understand. Please simplify it or, better, split it into two shorter sentences.

-          “… the power-law distribution is characterized by the so-called scale invariance.” – “the so called” should be deleted.

Comments for author File: Comments.pdf

Pretty good in general but can be improved.

Author Response

Comments and Suggestions for Authors

 

General observations

 

1. One may wonder what is the reason the manuscript was proposed to the journal Remote Sensing and not to a volcanology-specialized journal? This is not criticism!

 

The main reason is that it belongs to an INGV special issue.

 

2. The manuscript follows previous works of the same authors and other authors, correctly mentioned and cited, whose published data are incorporated in a much larger (numerically and phenomenologically) data set completed and updated to the end of June 2023 (!) by the authors themselves allowing a statistically more saturated, hence more precise and relevant interpretation of all available data (old and new) related to the eruptive history of Stromboli volcano from 1879 onwards. Large data set analysis and processing benefit from adequate statistical tools used by the authors. As so, the paper is a welcome and valuable contribution to the knowledge of eruptive activity of the volcano, its subtle time-dependent variation allowing better hazard assessments to be obtained in the future.

 

3. English language expression of the text is at a pretty good level (as far as a non-native English-speaking reviewer can appreciate it), but same language editing would be welcome.

 

We have asked our institutional English mother-tongue reviewer to correct the English style before re-submitting these corrections.

 

4. My only major critical observation is related to manuscript organization in chapters and subchapters which I found a bit illogical hence it could be improved. For example:

 

-         Chapter 2 Data and methods: Subchapters 2.1 Data and 2.2 Methods should be presented in reverse order, i.e., Methods first, then Data

 

Your observation is logical, given that normally the Methods chapter precedes the Data chapter. We have now separated Data from Methods, but were unable to reverse the order because we needed to present at first the data used in this paper and the way we collected them, and then describe the method used to carry out the statistical analyses.

 

-         Both subchapters 2.1. and 2.2 contain data, including diagram representation of data (e.g., Fig. 6 in Methods)

 

-         Data interpretation is introduced in the subchapter (2.1.) of data presentation (e.g., following Table 1b in Page 5): “The main cause of the non-uniform distribution of the events at Stromboli is …”. There are more such cases of data interpretation in 2.1. Data

 

Therefore, I kindly suggest the re-organization of the chapters/subchapters according to the following scheme (or other similar, in which methods, data and interpretation of data are clearly separated):

 

1. Introduction

 

2. Methods (including data acquisition sources and procedures)

 

3. Results

 

3.1. Data (processing and presentation in Tables and diagrams, including Numerical results)

 

3.2. Interpretation

 

3.2.1 Statistical models for the 1879-2023 data set

 

                       3.2.2 Statistical models for the 1985-2023 dataset

 

                       3.2.3 Statistical interpretation of results

 

4. Discussion and Concluding Remarks

 

 

 

5. Since my expertise does not cover the mathematical expression of statistical data processing, this methodological subject is not addressed at all in this review.

 

 

 

Specific observations (the text parts to which they refer to are highlighted with yellow in the pdf file)

 

1. Introduction

 

-         “…increasing more than ten times during the summer season” – please add the reason for that (e.g., visitors)

 

We have added "due to tourists" to clarify.

 

-         Fig. 1 – please mark with a well-visible symbol on the image in b) the actual location of Semaforo Labronzo

 

Thanks a lot. We have added a white square to show the position of the observatory.

 

-         “… mild Strombolian explosions that occur at a frequency of one every few minutes” – as I know the frequency of this kind of eruptions ranges from a few minutes to a few tens of minutes (witnessed by myself when visited the volcano in 2017)

 

We have changed the sentence accordingly.

 

-         “… this latest event of 19 July 2020 being at the boundary between ME and P…” – how that “boundary” was - or can be - defined (by numbers? else?)

 

We have now added the reasons: " because characterized by several discrete pulses involving more than one crater zone, and on the basis of seismic trace, VLP size, area involved by the fallout, height of the eruptive column and magma source depth".

 

-         What “magma fingering” means? Intrusion? Readers interested in Remote sensing are not familiar with this expression for sure. So, please explain it in a few words in brackets where first used.

 

Corrected.

 

2. Data and methods

 

-         “The details of the monitoring cameras are deeply illustrated by…”. What “deeply illustrated” means? How the details of the monitoring cameras can be “deeply illustrated”?

 

We have changed this sentence to: " The features of the monitoring cameras and their frequency of acquisition are described in detail by ".

 

-         “40 are uncertain major explosions (U), which means that they lie at the boundary between the persistent explosive activity and major explosions”. The same issue as above: how the “boundary” can be defined?

 

We have modified this sentence to clarify: " which means that they display a seismic signal (or "VLP size") at the boundary between the persistent explosive activity and major explosions (Giudicepietro et al., 2019, 2022; Calvari et al., 2021)."

 

-         Tables should not be divided in two separate pages (the case of Table 1b divided in pages 4 and 5)

 

Thanks a lot. We will check during the final pagination.

 

3. Numerical results

 

-         “… we consider the reduced and more recent portion of the catalogue…” – better shortened instead of reduced. If replaced, please do that in the whole text.

 

Modified throughout the text.

 

-         “… pointed out to an increasing number of …” – more correct: “… pointed out an increasing number of …”

 

Corrected.

 

-         “modifications of the shallow conduit” – “plumbing system” is a more appropriate term

 

We prefer to keep "shallow conduit" because it refers to the uppermost 200 m of the feeder system.

 

3.1 Statistical models for the 1879-2023 data set

 

-         Tables 3a (explosive events) and 3b (effusive events) can be joined as a single Table 3. If so, references in the text should be adapted to this change; the same for Tables 4, a, b, c, d, and for Figs 5a, b, c.

 

We have joined many of the tables: Tab 3, 4, 5, 6 and 7.

 

-         “… there is a slight advantage…” – what “advantage“ means in this context? Unclear.

 

Yes, since it was unclear we dropped this sentence, because not necessary for the purpose of the paper.

 

 

-         “focus” instead of “concentrate”

 

Thanks, corrected.

 

-         “… the graphic behaviour of the corresponding power-law models are shown … 1) behavior is correct grammatically; 2) is “behavior” the adequate word here? “graphic behavior” sounds strange to me; 3) “is” instead of “are”

 

Corrected. We have used "pattern" instead

 

-         “The blue star symbol in the abscissa…” in caption of Fig. 8: the blue star in the figure itself is too small and difficult to be seen. Can you make it bigger?

 

Done.

 

-         “…are reported in Figure 10a to 10c, for explosive, F effusive and FF+FFF effusive, respectively.” – the word “events” has to be inserted between “effusive” and “,respectively”

 

Corrected.

 

 

3.2   Statistical models for the 1985-2023 dataset

 

-         “… the results are summarized in Figure 11a to 11c and in Table 6a to 6c…” – commonly, Tables are placed first and the figures showing the data in a graphical form, are placed after them. This observation is valid for the whole manuscript.

 

Done.

 

-         Again, since the column heads are exactly the same, Tables 6 a, b, c and 7a, b, c can be combined as Table 6 and Table 7, respectively, having three rows each.

 

We have joined many of the tables: Tab 3, 4, 5, 6 and 7.

 

4.3   Statistical interpretation of results

 

-         “… as has also been experimentally demonstrated in this paper.” – please indicate where exactly in the paper that was demonstrated (intra-article reference)

 

We have changed the sentence in this way: “… the advantage of the power-law is that it is usually more accurate in modelling the events belonging to the tail of the distribution, as found in this paper (Figure 11).”

 

 

4. Discussion and Conclusive Remarks

 

-         “concluding”, instead of “conclusive”, sounds better in the chapter’s title

 

Thank you, corrected.

 

-         “… we present a new catalogue of events …” –Table SM1 displaying this catalogue has to be referred to

 

Added.

 

-         “… their occurrence is too low …” – do you mean the frequency of their occurrence is too low?

 

Yes, we have modified this sentence accordingly.

 

-         “ … the sizes and capacities of the plumbing systems of the two volcanoes are

 

-         very diverse …)” – I would suggest: “contrastingly different”, or just “different” instead of “very diverse”

 

We have modified this sentence accordingly.

 

 

-         “… a contribution to the risk assessment of both explosive and effusive events.” – Correct is “hazard assessment”, because “risk” is related to vulnerability, a subject not addressed in the manuscript

 

Corrected.

 

-         “Here, we propose and speculate …” - sounds a bit strange to me, but it is up to you to change

 

We have removed "and speculate".

 

-         “In addition, we have proposed to the well-known heavy tailed distributions ofexplosive activity, already taken into account by previous authors (Bevilacqua et al 2020b) for modelling the inter-event time for the major explosions and paroxysms recorded at Stromboli, the use of a power-law distribution.” – A quite complicated sentence difficult to follow and understand. Please simplify it or, better, split it into two shorter sentences.

 

Yes, we agree. We have changed the sentence in this way: “ Furthermore, for estimating the hazard related to the explosive activity at Stromboli, we have considered, in addition to well-known heavy tailed distributions such as the Lognormal, already taken into account by previous authors (Bevilacqua et al 2020b), the use of the power-law distribution.

 

-         “… the power-law distribution is characterized by the so-called scale invariance.” – “the so called” should be deleted.

 

Thanks, deleted.

 

Comments on the Quality of English Language

 

Pretty good in general but can be improved.

 

We have asked our mother tongue English reviewer to revise the text.