Skip to content

A ConvLSTM based deep neural network for volcanic clouds monitoring from space

Federica Torrisi 1, Claudia Corradino1, Ciro Del Negro1

  • Affiliations: 1Istituto Nazionale di Geofisica e Vulcanologia, Osservatorio Etneo -- Sezione di Catania, Piazza Roma 2, 95125 Catania, Italy

  • Presentation type: Talk

  • Presentation time: Thursday 09:00 - 09:15, Room R380

  • Programme No: 3.1.13

  • Theme 3 > Session 1

Abstract

Active volcanoes present significant hazards at various geographic scales. Volcanic clouds produced during explosive eruptions affect human health, climate, and aviation safety. Therefore, accurately capturing the temporal evolution of volcanic clouds is critical for understanding their dynamics and improving predictive capabilities. Given the highly dynamic nature of explosive eruptions, volcanic clouds can form, expand, and disperse rapidly. Therefore, high-temporal-resolution geostationary satellite data, such as that from MSG-SEVIRI (Meteosat Second Generation - Spinning Enhanced Visible and InfraRed Imager), is indispensable for near-real-time monitoring of these events. By capturing the swift changes in cloud formation and dispersion, we can identify patterns in the cloud's evolution and composition. Here, we propose a novel approach using a convolutional long short-term memory (ConvLSTM) model, a type of recurrent neural network (RNN), to track the spread of volcanic clouds using satellite data. A dataset of Ash RGB images, captured every 5--15 minutes from native SEVIRI data, was compiled from volcanic eruptions at Mt. Etna (Italy). We show that ConvLSTM models are able to solve complex spatiotemporal problems, effectively segmenting volcanic clouds and tracking their dispersion. This model provides timely and reliable information that supports aviation safety, emergency response, and public health monitoring, thereby enhancing decision-making during volcanic crises.