Constructing a New Catalogue of Greenland's Iceberg Calving Events through Seismic Data Analysis and Machine Learning

The Greenland ice sheet, a critical component of the global climate system, has played a substantial role in rising sea level, marked by a fourfold increase in mass loss due to iceberg calving between 1992-2000 and 2000-2011. Through the quantification of the spatio-temporal changes in Greenland’s ice mass loss resulting from iceberg calving, we gain a deeper understanding of the impacts of climate change.

The mass loss related to calving icebergs can be estimated by combining mechanical simulation of iceberg calving and inversion of seismic data. Seismic signals are generated by the time-varying force produced during iceberg calving on marine-terminating glacier termini. These events, known as glacial earthquakes, are recorded by the Greenland Ice Sheet Monitoring Network at tens of kilometres from the source.

However, differentiating these signals from tectonic events, anthropogenic noise, and other natural noise is challenging due to their complex frequency content (1-100s), multi-phase waveforms and low amplitude. To overcome this difficulty, we use a detection algorithm based on the Short-Time Average over Long-Time Average (STA/LTA) method and combine it with machine learning (Random Forests). By training the machine learning algorithm on seismic event catalogues containing more than 400 earthquakes and glacial earthquakes each, our approach is apt for identifying glacial earthquakes. Applying this methodology to continuous data offers the possibility to uncover smaller and previously undetected events. As a result, we present a comprehensive catalogue spanning several years and discuss its relevance and reliability. The generated catalogue allows us to develop new methods to better understand the spatio-temporal evolution of the ice-calving activity in the region. Among these, we will initially focus on locating and inverting the force of the largest events, providing a basis for testing new machine learning approaches for the characterisation of the source. This includes extracting properties like the iceberg volume and shape from both large and smaller events, ultimately advancing our understanding of Greenland's ice mass loss dynamics.