GNGTS 2022 - Atti del 40° Convegno Nazionale

GNGTS 2022 Sessione 2.1 211 NESTOREV1.0: A MACHINE LEARNING-BASED MATLAB PACKAGE FOR FORECASTING STRONG AFTERSHOCKS IN SEISMIC CLUSTERS S. Gentili 1 , P. Brondi 1 , R. Di Giovambattista 2 1 National Institute of Oceanography and Applied Geophysics - OGS, Udine, Italy 2 Istituto Nazionale di Geofisica e Vulcanologia, Roma, Italy NESTORE (Next STrOng Related Earthquake) is an algorithm for probabilistic forecasting of clusters in which a strong mainshock is followed by at least one aftershock of similar magnitude (greater than or equal to that of the mainshock minus one). Some previous versions of the algorithm have been successfully applied to California and Italian seismicity (Gentili and Di Giovambattista 2017, 2020, 2022). The algorithm analyzes selected seismicity features at increasing time intervals after mainshock occurrence, through a multiparameter pattern recognition approach based on decision trees. In particular, the analyzed features are related to the number of earthquakes, their energy, and spatiotemporal distribution, and are used to provide forecasting of the probability of strong subsequent events for the current seismic cluster. The main problems in this type of application are essentially two. The first is the number of available examples (clusters) which are generally on the order of tens, due to the limited time in which earthquake catalogs have high quality (maximum 50 years). In machine learning applications hundreds or thousands of examples are generally needed. The second is class imbalance, because a strong following earthquake is recorded only in a low percentage of the observed clusters. During these last years, the algorithm has been improved to increase its generalization capability despite the small number of clusters available in seismological databases and the presence of some outliers. A renewal of the old version of the software, both in terms of algorithm and usability, has been developed within the project “Analysis of seismic sequences for strong aftershock forecasting” funded by a grant of the Ministry of Foreign Affairs and International Cooperation in the framework of the scientific and technological collaboration between Italy and Japan. The current version of the software, NESTOREv1.0, is now mature enough to be distributed to the scientific community to be applied and tested in new areas. The software, which will be available on GitHub in the coming months, has a GUI (Graphical User Interface) or can be launched directly from the MATLAB command line. In this presentation, we will show the structure of the software and some examples of software outputs. In a related poster (Brondi et al. , 2022) we will show an example of the application of the new version of the algorithm to Italian seismicity. Acknowledgements. Funded by a grant from the Italian Ministry of Foreign Affairs and International Cooperation. References Gentili S. and Di Giovambattista R.; 2017: Pattern recognition approach to the subsequent event of damaging earthquakes in Italy. Physics of the Earth and Planetary Interiors, 266 , pp. 1-17. Gentili S. and Di Giovambattista R.; 2020: Forecasting strong aftershocks in earthquake clusters from northeastern Italy and western Slovenia. Physics of the Earth and Planetary Interiors, 303 , 106483. Gentili, S. and Di Giovambattista, R.; 2022: Forecasting strong subsequent earthquakes in California clusters by machine learning , Physics of the Earth and Planetary Interiors, accepted. Brondi, P., Gentili, S. and Di Giovambattista, R.; 2022: NESTOREv1.0 application to Italian seismicity Submitted to GNGTS 2022.