Earthquake monitoring is vital for understanding the physics of earthquakes and assessing seismic hazards. A standard monitoring workflow includes the interrelated and interdependent tasks of phase picking, association, and location. Although deep learning methods have been successfully applied to earthquake monitoring, they mostly address the tasks separately and ignore the geographic relationships among stations. Here, we propose a graph neural network that operates directly on multi-station seismic data and achieves simultaneous phase picking, association, and location. Particularly, the inter-station and inter-task physical relationships are informed in the network architecture to promote accuracy, interpretability, and physical consistency among cross-station and cross-task predictions. When applied to data from the Ridgecrest region and Japan, this method showed superior performance over previous deep learning-based phase-picking and localization methods. Overall, our study provides a prototype self-consistent all-in-one system of simultaneous seismic phase picking, association, and location, which has the potential for next-generation automated earthquake monitoring.
15 January 2024 in IEEE Transactions on Geoscience and Remote Sensing
In seismology, while training a specific deep learning model for each task is common, it often faces challenges such as the scarcity of labeled data and limited regional generalization. Addressing these issues, we introduce SeisCLIP: a foundation model for seismology, leveraging contrastive learning during pretraining on multimodal data of seismic waveform spectra and the corresponding local and global event information. SeisCLIP consists of a transformer-based spectrum encoder and an multilayer perceptron (MLP)-based information encoder that are jointly pre-trained on massive data. During pretraining, contrastive learning aims to enhance representations by training two encoders to bring the corresponding waveform spectra and event information closer in the feature space, while distancing uncorrelated pairs. Remarkably, the pre-trained spectrum encoder offers versatile features, enabling its application across diverse tasks and regions. Thus, it requires only modest datasets for fine-tuning to specific downstream tasks. Our evaluations demonstrate SeisCLIP’s superior performance over baseline methods in tasks like event classification, localization, and focal mechanism analysis, even when using distinct datasets from various regions. In essence, SeisCLIP emerges as a promising foundational model for seismology, potentially revolutionizing foundation-model-based research in the domain.
28 November 2023 in Geophysical Journal International
Discrimination between tectonic earthquakes (EQs) and quarry blasts is important for accurate EQ cataloguing and seismic hazard analysis. However, reliable classification of these two types of seismic events is challenging with no prior knowledge of source parameters. Here, we applied deep learning to perform this classification task in southern California and eastern Kentucky. Since the two regions differ significantly in available labelled data, class imbalance and waveform characteristics, we adopted different strategies for them. We directly trained a convolutional neural network (CNN) for southern California due to its data abundancy. To alleviate the class imbalance, the blast data were augmented by repeated sampling. The model for California yields F1-scores of >83.5 per cent when estimated by individual stations and >98.1 per cent by network average (i.e. averaging the CNN’s outputs on all available stations for each event). As eastern Kentucky has a much smaller data size, we apply transfer learning to the pre-trained California model to fit the Kentucky data. The transfer-learned model yields F1-scores of >86.9 per cent when estimated by individual stations and >96.7 per cent by network average. The transfer-learned model outperforms the model re-trained from scratch for the Kentucky data. Gradient-weighted class activation mapping shows the S onset and the S long-period coda are important to identify EQs and blasts, respectively. By visual inspections of the seismograms, the source locations, the origin time and the P-wave polarities, we verified that most of the events falsely predicted by our models are actually mislabelled by seismic analysts. Our models thus show great potential in helping seismic analysts find those mislabelled events which remain hidden in the historical catalogue. Our results demonstrate that deep learning can achieve high accuracy in seismic event discrimination and that transfer learning is effective and efficient to generalize deep learning models across different regions.
Seismic waves preceding the core phase PKIKP, known as PKP precursors, have been used to map small scatterers near the core‐mantle boundary (CMB). However, their low signal‐to‐noise ratio has required manual identification, hindering systematic and comprehensive analysis. Here, we develop GraphCursor, a model based on Graph Neural Networks (GNNs), to detect PKP precursors across multiple stations. Trained on manually picked precursors, GraphCursor identifies 37,512 precursor waveforms from 5,499 global earthquakes (2000–2024). We locate scatterers by stacking multistation probability functions and estimate their strengths relative to global averages. Our exhaustive search produces unprecedentedly abundant scatterers, including new discoveries of those beneath Antarctica. The distribution of most global scatterers largely correlates with ultra‐low velocity zones (ULVZs). GraphCursor's potential extends to identifying other deep seismic phases and aids in promoting the discovery of new deep Earth structures.
28 April 2026 in Earth and Planetary Science Letters
Continual learning needs to overcome catastrophic forgetting of the past. Memory replay of representative old training samples has been shown as an effective solution, and achieves the state-of-the-art (SOTA) performance. However, existing work is mainly built on a small memory buffer containing a few original data, which cannot fully characterize the old data distribution. In this work, we propose memory replay with data compression (MRDC) to reduce the storage cost of old training samples and thus increase their amount that can be stored in the memory buffer. Observing that the trade-off between the quality and quantity of compressed data is highly nontrivial for the efficacy of memory replay, we propose a novel method based on determinantal point processes (DPPs) to efficiently determine an appropriate compression quality for currently-arrived training samples. In this way, using a naive data compression algorithm with a properly selected quality can largely boost recent strong baselines by saving more compressed data in a limited storage space. We extensively validate this across several benchmarks of class-incremental learning and in a realistic scenario of object detection for autonomous driving.