Volcanic gas changes prior to Stromboli's major explosions are statistically significant

The generally mild activity of mafic, open-vent volcanoes is punctuated by the periodic occurrence of sudden, larger-than-normal explosive eruptions. Examples of these sudden mafic explosive events are the so-called “major explosions” that occur (2 to 4 times a year on average) at Stromboli volcano in Italy. These relatively small explosions (Volcanic Explosivity Index, VEI < 1) occur without no obvious precursory change in surface activity, and therefore pose a threat to volcanologists/monitoring staff, and to the population in the most severe cases. Past work has found a link between periods of high CO₂/SO₂ ratios in the volcanic gas plume and the occurrence of such explosions, but this association has never been statistically verified. Here, we report on nearly continuous observations of volcanic gas plume CO₂ and SO₂ concentrations and ratios in a diluted plume, as well as SO₂ flux measurements, recorded in a four year (2020 to 2023) activity period of Stromboli volcano, during which 22 such major explosions took place. Using the Conditional Logit model and the Receiver Operating Characteristic (ROC) analysis, we establish a statistical association between the occurrence of major explosions and periods of reduced SO₂ concentrations (very significant) and fluxes, and high CO₂/SO₂ ratios (significant), in the volcanic plume. These findings are interpreted in light of a simplified conceptual model that explains major explosions as caused by gas bubble accumulation at a rheological discontinuity, resulting from deceleration of the shallow convecting magma that supplies the “regular” Strombolian activity. Using results of statistical analysis, we develop a volcanic gas-based Composite indicator that successfully forecasts (by a-posteriori analysis) 71 % of the events on timescales of week(s). However, we find that this Composite indicator is associated with a large (32 %) False Positive rate and hence low precision (20 %). The significant role of SO₂ concentrations in the model indicates that other factors, such as plume direction and inter-crater variations in gas composition, may need to be taken into account for improving the forecasting performance of the method. Thus, while our results emphasize the importance of gas plume observations in volcano monitoring, they also highlight their current limitations as eruption forecasting tools.