Sub-surface processes

From EuroVolc Wiki
Jump to navigation Jump to search

Theme 3: Sub-surface processes description

Early identification of signs that magma is moving towards the surface, timing of eruption onset, possible transitions in eruptive style, and eruption end are all key for hazard monitoring and management. Volcano observatories carry out real-time monitoring of seismic, deformation and volatile emissions, which in the long term, can indicate charging of a magmatic system, and in the short term can track migration of magma toward the surface prior to eruption. Petrological data also provides valuable insights into sub-surface processes.

This theme will investigate integrated and automated modelling approaches for interpreting monitoring data (including petrological) together in near-real time to gain a better understanding of the sub-surface processes.

The Sub-surface processes theme is centred around research and collaboration on the following main topics:

  1. Rock-magma interaction: geophysical, geochemical and petrological signals of magma migration, chemical interaction between magma and surrounding rock
  2. Multidisciplinary assimilation of data: detection and understanding of the magmatic plumbing system through the combination of geophysical, geochemical and petrological data

Activities

Sub-surface processes activities involve:

  • Networking observations of sub-surface processes in relation to unrest and eruptions including: seismic, geodetic, magnetic, electrical, direct observational data, photos, videos and rock sample sets, geochemical, petrological, and experimental as well as GPS geodesy and levelling.
  • Workshop on multidisciplinary collaboration and integration
  • Initiating access to multidisciplinary observations from the Krafla Volcano Laboratory
  • Volcano pre-eruptive detection schemes: development of automatic algorithms for correlating real-time analysis and seismic and infrasound signals
  • Integrated modelling of pre-eruption data
  • Integration of petrological and real-time monitoring data
  • Physical and virtual access to research infrastructures and tools including the virtual diffusion modelling centre, seismic full wavefield modelling tools and remote sensing and numerical models and simulations

Work packages

The sub-surface processes theme is primarily the subject of the following Work Packages:

Outputs

Pre-eruptive unrest detection tools

Early identification that magma is moving towards the surface is very important for mitigation of volcanic risk, and detection of pre-eruptive unrest is underpinned by continuous real-time monitoring and real-time evaluation of these data. In principle joint real-time analysis and correlation of multiparameter data can pinpoint pre-eruptive unrest. However, given the potential unrest timescales, from years to minutes prior to an eruption, meaningful automatic multi-parameter pre-eruptive unrest detection is far from trivial in practice.

A python-based software tool, RETREAT, has been successfully developed that uses seismic array data and array processing techniques to help detect, quantify and locate volcanic tremor signals in real-time. The tool has been tested, in an academic environment, on both real time and archived data. The tool is now ready for testing and implementation in a volcano monitoring setting at observatories, and is also freely available to download, as a EUROVOLC community tool. More details on RETREAT can be found on the Open software page.

In addition, a second complementary real-time tool has been developed specifically to detect and track microseismicity at the planned small aperture HEIKSISZ array at Hekla volcano in Iceland. This tool has a slightly different focus to RETREAT, and uses an iterative approach to detect phase arrivals as well as using beamforming to locate signals. Ongoing development in the near future may allow the two tools to be unified and benefit from each other.

A secondary focus is on tracking seismic velocity changes associated with volcanic eruptions. This has been done using (1) coda-wave interferometry (CWI) on seismic multiplets, (2) analysis of ambient seismic noise, as well as (3) by analysis of an improved earthquake catalogue by using variations in Vp/Vs ratios.

Further details on these pre-eruptive unrest detection tools can be found in the deliverable report.

Rapid joint application of Sentinel-1 and GNSS data

Geodetic measurements are an important tool that volcano observatories have at their disposal to study volcanoes. To be truly useful to monitor pre-eruptive volcanic unrest progress needed to be made regarding the acquisition, processing and analysis of satellite based geodetic data.

We are working to make InSAR data a real-time volcano monitoring tool. To carry out joint rapid inversion of InSAR and GNSS, first we need to rapidly process the InSAR data. We achieve this using the LiCSAR system that provides automatically processed interferograms of Sentinel-1 images, and into which we have integrated the coverage of most European volcanoes. To analyse these data we have developed a machine learning algorithm to automatically detect new deformation in the InSAR data. Testing our approach on real data from an episode of pre-eruptive volcanic unrest shows the capability to detect changes in deformation and flag them as a potential sign of an impending eruption. Finally, we have developed a software, GBIS, to rapidly and jointly invert InSAR and GNSS data using Bayesian inference and estimate properties of sources of volcanic deformation, which we have made freely available online.

More details on this can be found in the deliverable report.

Machine learning tool kit and database

The recent success of deep learning approaches in a variety of pattern recognition fields (computer vision, natural language processing, speech recognition, game playing) and the growing availability of historical data in volcanology have posed the basis for the application of such techniques to automatic identification of patterns that lead to volcanic phenomena. However, the main limitation in investigating this line of approaches lies in the lack of annotations that describe the nature of the recorded data, preventing the use of standard supervised machine learning approaches.

We have developed a tool that performs anomaly detection in volcanic historical data in an unsupervised way, leveraging the hypothesis that normal activity dominates the event distribution and can therefore be treated as the main source of information for training a model able to identify deviations as anomalies. These anomalies could be attributed to changes in the volcano state. We validated the effectiveness of the proposed approach on data series from Etna and Stromboli, with promising results that confirm how this approach is able to identify anomalies in the data, often with a significant anticipation period from volcanic event occurrences.

The VOLUND (VOLcano UNrest Detection) toolkit is openly accessible through a GitHub repository at https://github.com/EUROVOLC-ML/VolUnD-Toolkit. Further information on the toolkit can also be found on the Open software page.

For more details, please refer to the deliverable report

Collaborations and ongoing activities