Key research themes
1. How do time-predictable models contribute to long-term earthquake prediction based on interevent times and magnitudes?
This research theme explores the validation and application of the time-predictable model, which posits that the interval between large earthquakes depends on the magnitude of the preceding mainshock. It matters because it offers a deterministic framework that can quantify the timing and magnitude of future large earthquakes over long-term horizons, improving probabilistic hazard forecasting based on empirical seismic data.
2. How can machine learning techniques, particularly neural networks and ensemble classifiers, improve short-term earthquake prediction and magnitude estimation?
This theme reviews the growing application of machine learning (ML) methodologies—including artificial neural networks (ANNs), recurrent neural networks (RNNs), convolutional neural networks (CNNs), and ensemble classifiers such as Random Forest and XGBoost—to analyze seismic time series and precursor signals for short-term earthquake forecasting and magnitude estimation. It reflects the transition towards data-driven approaches leveraging large seismic datasets, aiming to enhance prediction accuracy and lead-time over traditional statistical models.
3. What novel statistical and data-driven approaches enhance the operational forecasting of earthquakes beyond classical empirical laws?
With advances in machine learning, statistical seismology, and geodetic monitoring, recent research seeks to improve earthquake forecasting through combining improved empirical models with machine learning methods and geodetic data. This theme includes methods such as the EEPAS model describing precursory seismicity rate increases, spatial statistics revealing latent fault structures, and cross-disciplinary use of gravitational field variations and geodetic surface displacements. The goal is to improve reliability and lead time of forecasts driven by complex spatiotemporal seismic patterns.