Volcanic eruptions threaten neighboring populations. During an eruption, decision-making relies on event trees to evaluate different scenarios by comparing values of monitoring parameters with predefined thresholds. The lava flux — volume of lava emitted per unit of time — is one of the most useful, but it is very difficult to measure.

Our method aims at measuring lava volumes in near real-time from satellite radar images. Radar operates day and night, even through clouds or volcanic plumes. However, radar images contain distortions that make their interpretation challenging. Therefore, reconstructing lava bodies in 3D from radar images requires innovative tools. In this study we use Artificial Intelligence to perform this task.

We validate the method on the French volcanoes Piton de la Fournaise and La Soufrière de Guadeloupe by showing its capacity to reconstruct various volcanic structures. Then, we apply the method to two recent eruptions at La Soufrière, St. Vincent, in 2021, and at Shiveluch, Russia, in 2023.

This method shows great potential for estimating lava volumes from space, even with a very limited number of observations. It provides an efficient tool for timely hazard assessment, which could have operational applications in the context of crisis management of future eruptions.