| 10 Apr 2026
Sensitivity analysis and scaling of tsunamis generated by granular flows at Stromboli volcano: A numerical modeling approach
Abstract. Tsunamis generated by gravitational flows at volcanic islands pose a significant hazard, yet the sensitivity of wave heights to key source parameters remains poorly constrained. Using the non-hydrostatic multilayer HySEA model, we perform an extensive parametric study to assess the sensitivity of granular flow-generated tsunami wave heights at Stromboli volcano to six physical parameters: slide volume, initial submergence depth (elevation), density, basal friction, water-coupling coefficient, and source azimuth. A variance-based Sobol sensitivity analysis reveals that for the combined dataset of subaerial and submarine flows, the initial elevation (contributing ∼61 % to output variance normalized to 100 %) and volume (∼22 %) are the dominant controls on maximum wave height. When analyzed separately, subaerial tsunamis are primarily controlled by volume (∼60 %), while submarine tsunamis show a balanced sensitivity to volume (∼35 %) and initial submergence depth (∼37 %). We identify distinct scaling relationships between maximum wave height and landslide volume: a linear one for submarine landslides, with an exponential decay in efficiency as a function of submergence depth; a logarithmic fit for subaerial landslides, where the efficiency of wave generation per unit volume decreases for larger events. These relationships, modulated by secondary parameters like friction, provide a quantitative framework for rapid tsunami hazard assessment. Our results demonstrate a crossover in hazard potential, where subaerial slides tend to produce larger tsunami waves for smaller volumes, while submarine slides can produce larger waves for larger volumes due to the absence of logarithmic saturation. These results constrain the scaling laws needed to quickly invert tsunami observations into source volumes (and viceversa) and improve probabilistic hazard assessments by identifying the parameters that dominate uncertainty at Stromboli and similar islands.
This manuscript looks at the Sobol sensitivity of wave height from granular flow generated tsunamis off stromboli to various parameters
It is generally well thought out and written.
It would be good to clarify exactly how the granular flows are initialised for a given elevation q and volume V. e.g. what is the size, shape, thickness of the inital flow? Do all the edge lengths scale with volume or just some. How wide is it? Is q the top of the flow or the middle?
Also, the appendix gives three values of delta but only one is mentioned? are the other 2 kept constant? Are the three scaled together? A bit more explanation of this would be good.
line 58 along the shoreline - does this mean away from SdF - what shoreline is meant here?
line 60 location
140-150 more info on initialisation of granular flow, also how many layers used for water?
Fig1 I would suggest combining 0&2 and 1&3 as they are always referred to together and just note that there are actual gauges at these locations
164 I would spell out what these are again to help the reader
213 I am not quite sure what is varying for these standard deviations - a little more explanation would be good
224 has H been defined yet?
270 increase rather than correlation?
272 maybe phrase as: because dispersive effects have time to come into play
279 integral of the waveform (singular) and I assume that this is integral of absolute value of waveform?
282 thus we present
286 give section numbers rather than second third etc
291 the contribution of parameter delta exhibits ... (the variability is in response to the parameter)
294 results for the subaerial cases alone ... is the most influential
296 - repetition of a couple of sentences
Fig 4 I recommend putting the parameters in large print along the top of this figure for ease of viewing. Also I was curious the difference in Z-delta for combine versus separate but I assume this is because generally subaerial a lot bigger (at least for smaller volumes)so there is a big difference through that - is that correct? It might be worth mentioning
324 friction angle (color) and landslide density (symbol) [then delete the ()]
325 Overall a wider spread of max sea surface elevation... observed for larger volumes due to ...
328 This is expected as the lowest friction angle corresponds to the weakest basal resistance...
331 In accordance with the...
333 on H than rho and so uncertainty about delta is likely to lead to ...
339-348 This should be in methodology. is N number of data points that number of simulations run? looping over delta and V? Eqn 1 or 2 depending on q?
353 a clearer nonlinear
356 - it is the slope of the max wave height not the max wave height itself
fig 5 are the same (rather than coincide)
373 most pronounced variation in wave height with m_f, ranging from ...
375 Higher coupling (larger m_f)
384 - true for gauge 4&5 but not so obvious in other locations
northern lateral side (blue)
Section 4.4 - not sure why s is now used which is basically just -q isn't it?
416 for V = rather than in V =
419 these two coefficients [remind us which ones they are]
420 the error in the linear approximation [It is appropriate to refer to it as error in this situation]
421 I would reverse order of sigma_log and sigma_ref to keep in line with other locations in the ms
436 ... flow that is initially at rest and do not prescribe its ...
444 relatively insentive to failure locations up the slope
446 - it would be interesting to give example speeds/thicknesses for subaerial flows at impact with water
453 the further the slide travels
456 of H to delta
513 as evidenced by the low sobol sensitivity of this parameter.
around 545 - should acknowledge that this is only really appropriate at certain gauges where the crossover accords within the expected parameter range - in other locations this happens outside of expected range (where extrapolating significantly so equations may not hold)
570 vice versa (two words - also in abstract)
620 of wave height