The research, led by Andrea Bevilacqua and colleagues at the Istituto Nazionale di Geofisica e Vulcanologia in Pisa, analyzed a total of 67 explosive events, comprising 43 major explosions and 24 paroxysms. These two categories have long been used by volcanologists to distinguish the more intense interruptions of Stromboli’s ordinary activity, but the boundary between them has always been somewhat fuzzy. The team’s new dataset suggests that the fuzziness is real: when the maximum distances reached by ballistic projectiles and the total areas affected are plotted across all events, the distributions are remarkably continuous, with no clean separation between major explosions and paroxysms in terms of how far their projectiles fly.

The numbers are striking. Between 12 and 14 percent of major explosions produce ballistics that travel beyond 1,000 meters from the vents, while 29 percent of paroxysms extend projectiles over 2,000 meters. At the other end of the scale, between 23 and 37 percent of major explosions never surpass 500 meters, and between 17 and 42 percent of paroxysms fail to reach 1,500 meters. The average affected area was about 69,000 square meters for major explosions and roughly five times larger, about 360,000 square meters, for paroxysms. The largest distance recorded for any major explosion in the dataset, around 1,200 meters plus uncertainty, occurred on 19 July 2020, an event that other studies have placed right at the boundary between the two categories. The farthest paroxysm projectile, reaching approximately 2,500 meters, was thrown during the 22 May 1919 eruption, which damaged buildings in the island’s village.

What makes this study methodologically interesting is how the team dealt with the messy, incomplete nature of the historical record. Many of the most vivid accounts of Stromboli’s explosions date from the late nineteenth and early twentieth centuries, written mostly in Italian or German, and they describe affected places using place names that have evolved over time or distances rounded to multiples of 100 meters. Rather than trying to pinpoint the coordinates of individual clasts, the researchers developed a simplified mapping scheme in which each event is represented by a circular proximal area around the Crater Terrace, where all active vents lie, plus up to three circular sectors with variable radius and width that envelope the areas reached by projectiles.

This geometry is not arbitrary. Circular sectors mirror, to first order, the envelope of trajectories of projectiles ejected radially from a central vent, and most pre-existing hazard maps were drawn as hand-drawn lobes or open half-ovals that approximate this shape anyway. Each event was assigned to one of three uncertainty classes depending on the quality of the available information: low uncertainty for well-mapped major explosions, moderate uncertainty for events fully described around the source, and high uncertainty for historical paroxysms with weak directional constraints. The uncertainty ranges, typically representing a mean enlargement of 50 to 100 meters on radial distances, were treated as independent uniformly distributed variables and propagated through the entire analysis using Monte Carlo simulations that randomly perturbed the simplified maps 1,000 times.

The directional results reveal a strong asymmetry that differs significantly between the two event types. About 87 percent of the sector bisectors for major explosions point into the eastern half-plane, with a pronounced probability peak of 77 percent, plus or minus 2 percent, toward the southeast at an azimuth of 140 degrees. Paroxysms, by contrast, show 64 percent of their bisectors directed into the northern half-plane, with a maximum probability of 70 percent, plus or minus 9 percent, toward the north. The researchers note that the most frequented trails and inhabited areas are also the easiest places to survey after an eruption, so these peak directions may partly reflect observation bias, a problem the team flags as needing dedicated mitigation methods in hazard map production.

The physical reasons for these directional differences likely lie in the complex interplay between the volcano’s conduit and crater architecture and the very different scales of the two eruption types. Paroxysms appear capable of reshaping the shallow conduits and craters more deeply than major explosions, and laboratory detonation experiments have shown that explosion depth and pre-existing crater morphology can strongly influence the resulting jets and projectile dispersal. Prevailing winds, which at Stromboli blow mainly toward the east and southeast, appear insufficient to explain the observed patterns on their own; a typical 5-meter-per-second crosswind would displace a 10-centimeter lithic clast by only about 60 to 125 meters during a major explosion. Bouncing and rolling of projectiles on the volcano’s steep slopes may also extend their final resting positions beyond the initial impact points.

The stakes of this work are high because ballistics are the most common cause of fatal accidents within 5 kilometers of active volcanoes worldwide, causing tens of deaths and hundreds of injuries in recent decades. At Stromboli specifically, the current early-warning systems can detect precursor signals only a few minutes before major explosions and paroxysms, far too little time for anyone on the summit trails to reach safety. Quantitative assessments of ballistic hazard at different sites of the island are therefore a crucial input for risk mitigation and emergency planning, particularly for a volcano that attracts large numbers of tourists to its slopes every year.

The researchers are careful to note the limitations of their reconstructions. They focused exclusively on the most dangerous projectiles, lithic and scoria clasts larger than about 5 to 10 centimeters in diameter, and did not quantify the areal density of projectiles, which can vary by up to two orders of magnitude with distance from the crater. Under-recording of projectiles landing in the inaccessible Sciara del Fuoco, the horseshoe-shaped depression on the volcano’s northwest flank, is likely significant for both event types, since any ballistics falling more than about 1,200 meters to the northwest would land in the sea and go undocumented unless directly observed. Despite these challenges, the team found that the influence of the mapping uncertainty on the reconstructed dispersal areas was limited, and the resulting statistics were robust.

This paper is the first part of a two-part contribution. The dataset of distances, directions, and affected areas, complete with quantified uncertainty, provides the essential foundation for the companion study, which presents the first probabilistic hazard maps of ballistic fallout at Stromboli. Together, the two papers transform a century and a half of scattered observations, from a 1915 conglomerate boulder photographed by early volcanologists to impact craters documented by drones after the 2022 major explosion, into a rigorous statistical framework that could ultimately help decide where it is safe to walk on one of the world’s most persistently active and most closely watched volcanoes.

Subject of Research: Ballistic projectile hazard mapping from major explosions and paroxysms at Stromboli volcano

Article Title: Ballistic projectile hazard of major explosions and paroxysms at Stromboli (Italy) with uncertainty quantification – Part 1: Mapping method and data analysis

Article References: Bevilacqua, A., Landi, P., Del Carlo, P., Neri, A., & Pompilio, M. (2026). Ballistic projectile hazard of major explosions and paroxysms at Stromboli (Italy) with uncertainty quantification – Part 1: Mapping method and data analysis. Natural Hazards and Earth System Sciences, 26(9), 4479-4502. https://doi.org/10.5194/nhess-26-4479-2026

Image Credits: AI Generated

DOI: 10.5194/nhess-26-4479-2026

Keywords: Stromboli, volcanic ballistics, paroxysms, major explosions, hazard mapping, uncertainty quantification, Monte Carlo simulation, volcanic hazard, Italy, INGV, Natural Hazards and Earth System Sciences, risk mitigation