COMPASS Student Sam Stockman and Statistics Institute member Dan Lawson have worked collaboratively with seismologist Max Werner from the School of Earth Sciences to publish a paper into the American Geophysical Union journal; Earth's Future. Titled 'Forecasting the 2016-2017 Central Apennines Earthquake Sequence with a Neural Point Process', the paper constructs a temporal point process model for forecasting earthquakes during aftershock sequences such as the one that struck central Italy in 2016. Through advances in the methodology surrounding earthquake data collection, datasets now contain significantly more earthquakes than in previous years since they now also include the hundreds of thousands of small magnitude events that comprise an earthquake sequence. The model that the paper introduces, is more scalable to such large datasets than the state-of-the-art, as well as more robust to dataset artifacts such as the missing data following particularly large quakes. If the model forecasts based on the newly exposed small earthquakes found in the dataset, then it outperforms the state-of-the-art. The model only forecasts the times and magnitudes of future earthquakes and so there is ongoing work in extend it to spatial forecasts.

The paper was recently featured in a nature article on modern methods of aftershock forecasting found here

Stockman, Lawson and Werner's paper can be found here