Is the volume-frequency distribution of eruptions a power-law? Accounting for volume uncertainty in modeling the size distribution of volcanic eruptions

Ferrara, Salvatore; Selva, Jacopo; Natale, Jacopo; Marzocchi, Warner

doi:10.5194/egusphere-2026-1415

Preprints

https://doi.org/10.5194/egusphere-2026-1415

Preprints

08 Apr 2026

| 08 Apr 2026

Status: this preprint is open for discussion and under review for Natural Hazards and Earth System Sciences (NHESS).

Is the volume-frequency distribution of eruptions a power-law? Accounting for volume uncertainty in modeling the size distribution of volcanic eruptions

Salvatore Ferrara, Jacopo Selva, Jacopo Natale, and Warner Marzocchi

Abstract. Forecasting the size of a future volcanic eruption in densely populated areas is a key aspect of volcanic hazard and risk assessment. Estimates of the next eruption's size for both long- and short-term forecasts are typically based on the sizes of past events. Using the erupted volume as a proxy for eruptive size, forecasts are often obtained from the sampling of a power-law distribution. Notwithstanding, the distribution of the measured/inferred erupted volume of past eruptions often appears markedly different from a power-law. Here, we consider how the uncertainty on the volume of past eruptions may affect the shape of a hypothetical power-law distribution. The goal is to understand if the distribution of real data is compatible with an Exponentially Modified Gaussian distribution (EMG) that includes both the power-law and the uncertainty on the observed volumes. We apply this method to two large high-risk calderas, Campi Flegrei, Italy, and Taupo, New Zealand, but it can be potentially applied to any volcano. We find that the EMG distribution provides a good statistical fit to both volcanoes' eruptive records, supporting the use of a power-law distribution for forecasting the volume of the next eruption.

Received: 16 Mar 2026 – Discussion started: 08 Apr 2026

Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims made in the text, published maps, institutional affiliations, or any other geographical representation in this paper. While Copernicus Publications makes every effort to include appropriate place names, the final responsibility lies with the authors. Views expressed in the text are those of the authors and do not necessarily reflect the views of the publisher.

Download & links

Salvatore Ferrara, Jacopo Selva, Jacopo Natale, and Warner Marzocchi

Status: open (extended)

Post a comment Subscribe to comment alert

RC1:
'Comment on egusphere-2026-1415', Anonymous Referee #1, 07 Jul 2026 reply
This manuscript presents an analysis of the distribution of the size of past eruptions, expressed in terms of magnitude, aimed at exploring whether a power law distribution for the erupted volumes holds for individual volcanoes. Two volcanoes are considered, namely, Campi Flegrei (Italy) and Taupo (New Zealand). The analysis is clear in its objectives and results: a power law for volumes (which translates into an exponential distribution when magnitudes are considered) appears to be consistent with the data when normally distributed volume uncertainties are added.
I tend to agree with the presented results. Previous work involving one of the co-authors (WM) shows that global volcanic eruptions display (above a certain threshold) a power law distribution of the erupted volumes. If that still holds for individual volcanoes is still an open problem. That is mainly due to the fact that commonly only a limited number of eruptions from each individual volcano is known in sufficient details to be included in statistical analyses. The two volcanoes considered here are among the best studied in the world, and among the few where statistical tests can be applied with some robustness. This work therefore adds an important piece of evidence to the overall puzzle: for two of the best studied volcanoes in the world, the eruption volume distribution is still a power law, as for the global distribution.
That said, I have a number of observations that I believe can contribute to improve the quality of the contribution.
First, the objective expressed in the title is not clear enough. The question “Is the volume-frequency distribution of eruptions a power-law?”, expressed in the title, should make clear reference to the distribution from individual volcanoes. That would eliminate false expectations by the reader on the content of the contribution, and more clearly anticipate its contents.

The authors state that (abstract) “forecasts are often obtained from the sampling of a power-law distribution”, but do not provide any example of that, which would be instead useful.

When referring to the work in Papale et al. (2021) (involving one of the co-authors of this contribution: WM), the manuscript states that eruption volumes have been modelled as being distributed according to a power law “by analogy” with other similar distributions such as the one in the Gutemberg-Richter law for earthquakes. Thai is incorrect. That paper analyses global volcanic eruption databases and finds that global volcanic eruptions are (above a threshold) distributed as a power law – it does not just model them as a power law “by analogy” with something else.

Immediately after (line 46), the manuscript says “Yet it is easy to show that even simple plots of observed eruption volumes sometimes appear to deviate significantly from a power-law”. The “simple plots” in that statement to are almost certainly referred to individual volcanoes – while the reference to Papale et al (2021) immediately before this statement refers to the global scale. As from point 1) above, this contribution appears to frequently mix up the analysis over the global scale with the one in this paper on individual volcanoes. However, the individual volcano scale does not necessarily mimic the global scale (or vice-versa). This manuscript shows that also the individual volcano scale seems to display power law distributions for the erupted volumes. But even if it didn’t, that would not necessarily imply that the global scale does not. The two global and local (individual volcano) scales should be more clearly separated throughout the whole text.

Line 78: the last part of the sentence: “which is also a function of their eruptive style” does not seem to add any relevant information while making the sentence less clear. I suggest to remove it.

Lines 95 and followings: the authors assume that the errors on volume estimates are log-normal distributed with mean on the nominal value. However, the examples they report for potential causes of bad volume estimates (line 85 and followings) all suggest possible dominance of underestimations. Would other distributions, e.g., skewed log-normal, be worth of consideration?

Lines 110-112: the discussion on Mmin here is unclear. The authors say the catalogues are incomplete, therefore, they select Mmin as the value above which the distribution appears to be of the exponential type. Then, if their EMG distribution (the exponential plus log-normally distributed error) fits the data well, they can use it to extend below Mmin to forecast down to the desired M. That discussion seems to ignore the fact that there must be an Mmin below which the distribution is not exponential anymore – which corresponds to a Vmin below which the volume distribution is not anymore of the power law type. Just assuming that they can extrapolate down to any desired M (or V) seems therefore unjustified.

Line 116: the sentence starting with “This volcanic field” is far too long and difficult to read.

Tables 1 and 2. The original data used to construct these Tables are deposit volume. However, neither the deposit volumes, nor the assumed average deposit densities which would allow to retrieve deposit volumes from the reported data, appear in the manuscript. That makes retrieving the original data impossible (the many citations in the table captions do not help: this manuscript should be self-sufficient and present the real data employed rather than their further elaborations, or at least allow the reader to retrieve the real data).

The two figures 1 and 3 are poorly explained, while more clarification seems necessary. Both figures show, for CF and Taupo respectively, the log-likelihood contour plots for the quantities b and Mmin. However, b depends on only Mmin and Mave (line 65). For a given dataset, e.g., the one from either Table 1 or Table 2, once Mmin is defined, Mave is also defined. Thus, it seems that there should be a univocal relationship between b and Mmin for each one of the two CF and Taupo cases, which is not the case in the plots of Fig. 1 and 3. I may misunderstand something here, but in all cases, more explanation would help.

Line 150 and following. The authors may want to clarify, here as well as when describing similar results for the Taupo case, that their first tests with the Lilliefors and AD methods do not involve convolution with a normal error distribution. That should be stated before saying that the corresponding null hypothesis is rejected (which at this stage seems to contradict the results anticipated in the abstract). I could understand that only lines below, after I had already re-read a few paragraphs trying to understand what I was missing in the arguments by the authors.

Lines 157-158. This point is connected to point 9 above. Here the authors say that to calculate magnitudes they start from the DRE and adopt some constant density to convert from mass to volume. However, again, DRE is not the primary data, rather, it is an estimate which is obtained by the deposit volumes – the primary data – after combining it with some assumptions on the average density of the deposit and average density of the pre-eruptive magma excluding gas bubbles. When the authors assume the DRE as the primary data and apply on it some average density to convert into mass (thus magnitude), they are not necessarily consistent with the original deposit volume data - it depends on what assumptions were originally made to estimate the DRE, something which is not said in this manuscript. My strong suggestion is that the authors refer, in the text as well as in Tables 1 and 2, to the original data represented by the deposit volumes, and state what densities they use to convert from the deposit volumes to the erupted masses (the DRE is even not necessary in this analysis).

Line 189: the Ouranui eruption from Taupo volcano is said to have produced “several hundred km3 of products” (citations provided). However, the LaMEVE database reports 1170 km3 of deposits (therefore it classifies that eruption as VEI 8). Could the authors explain such a discrepancy?

Line 208. The authors assume a density for the DRE at Taupo of 2.45 kg/m3; while that same density is assumed to be 2.4 kg/m3 for Campi Flegrei (line 158). In light of the expected significant variabilities concerning the many eruptions from the same volcano, such differences with no explanations provided sound like ad hoc choices decreasing the robustness of the arguments and of the findings. I recommend that more explanation is provided here.

Lines 237-238: the authors say: “In forecasting mode, 𝛽 can be estimated via MLE, while 𝑀𝑚𝑖𝑛 should be defined according to the application's needs.” However, there must be a physical limit in Vmin (thus in Mmin in the exponential distribution), as no power law distribution can continue to indefinitely small values; e.g., for the global distribution of volcanic eruption volumes, Vmin has been found to be close to 170 Mm3 (mean value) (Papale et al., 2021, cited in the manuscript). Stating that Mmin can be defined according to own needs seems therefore unjustified.

Lines 261-262: we are here in the Conclusions section, and the authors write that “the null hypothesis of the magnitude exponential distribution cannot be rejected by the observations”. I suggest to remind here that the magnitude exponential distribution corresponds to a power law volume distribution. That would be useful in general, and guarantee that the readers just interested in the overall conclusions correctly understand the results of this analysis.

Lines 264-266: there is previous literature 1) largely showing that also for volcanic eruptions as for other natural phenomena, low intensity events are more frequent, and 2) referring to continuous distributions to overcome the problems of deriving from discrete eruption classes. As they are presented at the indicated lines within the Conclusions section, these seem to be new findings or novel approaches from this contribution, but they are not. I recommend that this is more fairly stated, and proper references are included.

Lines 307-308 (Appendix): “For magnitudes higher than 5, the simulated samples are left continuous, as the catalogs do not present such discretization above this value”. It would be more correct to say that for such magnitudes the discretization takes place on a finer mesh. There is still discretization, as the deposit volume estimates are unfortunately still rough.

Reply
Citation: https://doi.org/10.5194/egusphere-2026-1415-RC1

Salvatore Ferrara, Jacopo Selva, Jacopo Natale, and Warner Marzocchi

Viewed

Total article views: 415 (including HTML, PDF, and XML)

HTML	PDF	XML	Total	BibTeX	EndNote
264	125	26	415	23	28

HTML: 264
PDF: 125
XML: 26
Total: 415
BibTeX: 23
EndNote: 28

Views and downloads (calculated since 08 Apr 2026)

Month	HTML	PDF	XML	Total
Apr 2026	154	48	17	219
May 2026	85	66	5	156
Jun 2026	20	8	4	32
Jul 2026	5	3	0	8

Cumulative views and downloads (calculated since 08 Apr 2026)

Month	HTML	PDF	XML	Total
Apr 2026	154	48	17	219
May 2026	85	66	5	156
Jun 2026	20	8	4	32
Jul 2026	5	3	0	8

Viewed (geographical distribution)

Total article views: 405 (including HTML, PDF, and XML) Thereof 405 with geography defined and 0 with unknown origin.

Country	#	Views	%

Latest update: 13 Jul 2026

Short summary

We develop a probabilistic framework for modeling volcanic eruption sizes. Assuming a power-law distribution on erupted volumes, we analyze how measurement error can cause the observed trend to deviate from such a distribution. We apply this framework to Campi Flegrei, Italy, and Taupo, New Zealand, and observe that when the error is properly accounted for, the data distribution is compatible with a power-law, supporting the use of such a distribution in size forecasting.


Total:	0
HTML:	0
PDF:	0
XML:	0