Related Experiment Video
Updated: Jul 16, 2026
09:44
Recording and Analyzing Multimodal Large-Scale Neuronal Ensemble Dynamics on CMOS-Integrated High-Density Microelectrode Array
Published on: March 8, 2024
Neural network models for earthquake magnitude prediction using multiple seismicity indicators
1Department of Civil and Environmental Engineering and Geodetic Science, The Ohio State University, Columbus, OH 43210, USA.
International Journal of Neural Systems
|March 30, 2007
Summary
This study uses neural networks to forecast major earthquake magnitudes using seismicity indicators. Recurrent neural networks demonstrated the highest accuracy in predicting seismic events.
Area of Science:
- Geophysics and seismology
- Artificial intelligence in geoscience
- Earthquake prediction and hazard assessment
Background:
- Seismic event prediction remains a significant challenge in geophysics.
- Existing methods lack established empirical relationships between seismicity indicators and future earthquakes.
- Short-term seismic hazard assessment requires advanced analytical approaches.
Purpose of the Study:
- To investigate the efficacy of neural networks for predicting the magnitude of the largest seismic event in the subsequent month.
- To compare the performance of three distinct neural network architectures: feed-forward Levenberg-Marquardt backpropagation (LMBP), recurrent neural networks (RNN), and radial basis function (RBF) networks.
- To evaluate prediction accuracy using robust statistical measures.
Main Methods:
- Analysis of eight seismicity indicators derived from Gutenberg-Richter and characteristic earthquake magnitude distributions.
- Modeling earthquake prediction using LMBP, RNN, and RBF neural networks.
- Validation of models using data from Southern California and the San Francisco Bay region, assessed by probability of detection, false alarm ratio, frequency bias, and true skill score.
Main Results:
- The recurrent neural network (RNN) model exhibited superior prediction accuracy compared to LMBP and RBF models.
- Performance evaluation across different statistical measures indicated the RNN's effectiveness in forecasting seismic event magnitudes.
- The study successfully applied machine learning to analyze complex seismicity patterns.
Conclusions:
- While high-certainty earthquake prediction is not yet achievable, this research offers a scientifically rigorous method for evaluating regional seismic hazard.
- Neural networks, particularly RNNs, show promise for improving short-term earthquake forecasting.
- The findings contribute to a more scientific approach to understanding and managing earthquake risks.