For centuries, the main rupture forecast on both the laboratory and field scale was considered as a question of the far future or just an unsolvable task. The main tool of description of the seismic process was a purely statistical approach that allowed forecasting more or less exactly the possible area and intensity, but not the expected time of the future seismic events. Last decades the progress in Machine Learning (ML) raised hope that a very significant progress can be achieved in this direction. ML allows analyzing complex time series containing many informative features of the prolonged fracture process, namely, sequences of micro-events, preceding the final rupture: for example, we know that foreshocks are forewarning many strong EQs. ML of the data sets previous to the main rupture allows predicting the future behavior of the system. Informative precursors observed before the main rupture on the laboratory and field scale are plenty. The decisive role of ML in EQ forecast is its huge potential to aggregate the whole multitude of precursory (learning) data with a store of special analyzing computer programs and train them to find regular diagnostic signs of approaching the critical state.
The problem addressed is the forecast/prediction of time (with different accuracy for forecast and prediction), magnitude and location of the moderate/strong main rupture events on the laboratory and field scale using different ML tools. To achieve this goal, one should analyse time series of different possible precursors (seismic, strains, hydraulic, magnetic, etc.) and assess the forecasting/prediction quality at approaching the critical state.
Manuscript submitted to this Research Topic should address, but are not limited to, the bellow points:
1. Application of different ML tools to "laboratory earthquakes" (stick-slip process);
2. Testing different ML tools to forecast moderate/strong earthquakes using seismic data;
3. Strong/moderate earthquake forecast using ML on a combination of different precursory processes: strain, weak seismicity, hydrodynamic, geomagnetic and other effects;
4. Strong/moderate earthquake forecast as the imbalanced data sets' problem;
5. The problem of optimization of precursory data (avoiding under- and overfitting);
6. Using nonlinear dynamics (complexity analysis) approach for strong earthquake forecast (space plots of seismic time series);
7. Machine learning for revealing synchronization and triggering of seismicity by weak forcings;
8. Review of existing earthquake forecast algorithms/competitions;
9. Advancing earthquake forecasting by machine learning of satellite data.
For centuries, the main rupture forecast on both the laboratory and field scale was considered as a question of the far future or just an unsolvable task. The main tool of description of the seismic process was a purely statistical approach that allowed forecasting more or less exactly the possible area and intensity, but not the expected time of the future seismic events. Last decades the progress in Machine Learning (ML) raised hope that a very significant progress can be achieved in this direction. ML allows analyzing complex time series containing many informative features of the prolonged fracture process, namely, sequences of micro-events, preceding the final rupture: for example, we know that foreshocks are forewarning many strong EQs. ML of the data sets previous to the main rupture allows predicting the future behavior of the system. Informative precursors observed before the main rupture on the laboratory and field scale are plenty. The decisive role of ML in EQ forecast is its huge potential to aggregate the whole multitude of precursory (learning) data with a store of special analyzing computer programs and train them to find regular diagnostic signs of approaching the critical state.
The problem addressed is the forecast/prediction of time (with different accuracy for forecast and prediction), magnitude and location of the moderate/strong main rupture events on the laboratory and field scale using different ML tools. To achieve this goal, one should analyse time series of different possible precursors (seismic, strains, hydraulic, magnetic, etc.) and assess the forecasting/prediction quality at approaching the critical state.
Manuscript submitted to this Research Topic should address, but are not limited to, the bellow points:
1. Application of different ML tools to "laboratory earthquakes" (stick-slip process);
2. Testing different ML tools to forecast moderate/strong earthquakes using seismic data;
3. Strong/moderate earthquake forecast using ML on a combination of different precursory processes: strain, weak seismicity, hydrodynamic, geomagnetic and other effects;
4. Strong/moderate earthquake forecast as the imbalanced data sets' problem;
5. The problem of optimization of precursory data (avoiding under- and overfitting);
6. Using nonlinear dynamics (complexity analysis) approach for strong earthquake forecast (space plots of seismic time series);
7. Machine learning for revealing synchronization and triggering of seismicity by weak forcings;
8. Review of existing earthquake forecast algorithms/competitions;
9. Advancing earthquake forecasting by machine learning of satellite data.
