Local Magnitude Estimation via an Attention-Based Machine Learning Model

Abstract Rapid and reliable earthquake magnitude estimation is crucial for disaster management, scientific research, and resource conservation across multiple fields, especially during the initial stages of event detection. The most reliable traditional methods rely on complete waveform records, including earthquake epicenter distance and waveform amplitude, which can delay magnitude assessment. Machine learning techniques offer a promising avenue for capturing nonlinear relationships within seismic data, enhancing both information extraction and timeliness in magnitude estimation. In this study, we introduce an Attention-based machine learning model for MAGnitude estimation (AMAG) tailored for real-time earthquake monitoring. Using two independent datasets for training and testing, the results demonstrate the efficacy of our method in accurately predicting earthquake magnitudes. The magnitude prediction errors on the two test sets are −0.2 and −0.1, respectively, and the picking errors for both are 0.02 s. Our approach can be used directly for different time windows and signal lengths (at least 1 s) without retraining. We investigate the influence of signal-to-noise ratio, distances, and the integration of attention mechanisms. The attention mechanism facilitates the identification of the first motion and provides insights into the network’s focus areas, thereby establishing a relationship between waveform characteristics and earthquake magnitude. In addition, we systematically explore the impact of network architectures, loss functions, and signal lengths on prediction performance. Our findings reveal that a network with a depth of four layers and a convolution kernel size of five yields optimal prediction accuracy, with mean square error identified as the most effective loss function. When the input waveform is six seconds long, with equal durations of noise and signal, the model’s prediction accuracy is optimized. Our study underscores the potential of machine learning-based magnitude estimation for real-time earthquake monitoring, offering novel opportunities to mitigate natural disaster impacts, minimize casualties, and safeguard lives and property.