Forecasting aftershocks after an earthquake could get more accurate with researchers at the University of California, Santa Cruz, and the Technical University of Munich creating a new model that uses deep learning. Deep learning is a type of machine learning based on artificial neural networks in which multiple layers of processing are used to extract progressively higher-level features from data.
The new model developed by these researchers is called the Recurrent Earthquake foreCAST (RECAST), which they have shown in a study to be more flexible and scalable than the earthquake forecasting model currently used, the Epidemic Type Aftershock Sequence (ETAS). The new model outperformed the ETAS model for earthquake catalogues, each with 10,000 events and more.
The researchers started with using the ETAS model to simulate an earthquake catalog, after which they tested the RECAST model using real data from the Southern California earthquake catalog. The team found the RECAST model performed better than the ETAS model at forecasting aftershocks, as the amount of data increased.
The computational effort and time were also significantly better for larger catalogs. Machine learning technology to forecast quakes and aftershocks was not ready till recently although this was not the first time they used it.
However, the RECAST model was found to be more accurate due to the new advances in machine learning. RECAST and its flexibility is now being seen as a threshold technology that would open up new possibilities in earthquake and aftershock forecasting—especially its ability to adapt to large data. The researchers say models that use deep learning could potentially incorporate information from multiple regions at once.