No abstract

• We stress a gelatin block by excavating its surface and imposing lateral strain, then observe air-filled cracks trajectories • We numerically simulate the trajectories and retrieve the stress state of the gelatin, 10 then we forecast new crack arrivals at the surface 11 • Our strategy correctly retrieves the ratio between lateral extension and excava-12 tion depth and leads to effective forecasts of eruptive vents

In many volcanic regions, eruptive vents may be spatially scattered: they sometimes cluster along rift zones or are distributed over areas that may exceed 10,000 km 2 . Some of these regions are densely populated. In order to better protect human life and infrastructure, it is important to better understand the factors determining vent distributions and improve vent location forecasts.The most common approaches to probabilistic forecasts of future vent opening locations rely on the spatial density of past eruptive events, sometimes complemented with the surface distribution of structural features, such as faults and fractures (Bevilacqua et al., 2015;Connor & Hill, 1995;Martin et al., 2004;Selva et al., 2012). Such models, however, often remain poorly constrained due to scarce or spatially sparse data and cannot be easily validated in volcanic systems where eruptions are infrequent.Recently, Rivalta et al. (2019) proposed a mechanical-statistical approach to inversely constrain the state of stress, and thus magma pathways, of a volcanic region on the basis of the known location of magma reservoirs and past eruptive vents. Dike trajectories are assumed to follow a "least resistance to opening" path calculated from the elastic stress field, which is optimized so that any magma batch released from the magma reservoir reaches one of the past eruptive vents. Once the stress field is constrained, the trajectories of future dikes can be forecast. Rivalta et al. (2019) applied the concept only to Campi Flegrei caldera in Italy, performing inversions on two stress parameters: namely, the tectonic and the unloading stress. As independent estimates of such parameters in nature are affected by large uncertainties, it remains unclear how accurately the model can capture them, how much other factors, such as medium layering, were biasing the results, and how this would affect the forecast.