The first work package aims to detect and classify the spatio-temporal patterns of volcano deformation using the rapidly growing global archive of satellite imagery. Existing machine learning techniques for analysing large datasets were developed on small datasets of ~30,000 images. We have now scaled up these proof-of-concept studies to analyse millions of images and have developed new machine learning methods that do not rely on pre-conceived models. Alongside this, we are developing new methods for separating different signals within time-series data, including separating deformation signals from atmospheric noise using deep learning, and separating signals from magmatic and hydrothermal systems. Finally, we have developed a framework to systematically compare the deformation patterns between volcanoes, and have applied it regionally.
The second goal is to understand the relationship between the long-term development of magmatic plumbing systems and the short term deformation observed in our satellite archive. Over timescales of hundreds of thousands of years, a hot, weak zone develops that flows in response to short-term pressure changes. Using numerical modelling, we found that relatively cold magma systems exhibit cycles of uplift and subsidence, while comparatively hot plumbing systems experience solely uplift. These predictions fit well with observations from magmatic systems where geophysical methods have been used to measure subsurface temperature and satellites have observed active deformation.
The third goal is to develop new modelling methods that incorporate our latest understanding of magmatic systems and can be linked to satellite observations. In the laboratory, we have developed scaled models of an inflating magma chamber (golden syrup) within an elastic crust (gelatin), where we can measure both the surface deformation and the pressure within the chamber. Our initial results show that fracturing around the magma chamber affects the rate of surface uplift, a process that is not considered in existing numerical models. We are now developing the next generation of numerical models that combines solid particles with fluid dynamics, allowing us to understand the effect of crack formation and magma flow on surface deformation.
Finally, we are combining our expertise to respond to an earthquake swarm in Ethiopia, where a 50 km long magmatic dyke has formed between the volcanoes, Fentale and Dofen. The high-quality satellite data from the CosmoSkyMed satellite reveal the evolution of the magmatic intrusion and surface fracturing in unprecedented detail and will be a major focus for research in the second half of the project.
