We here present one lightweight phase picking network (LPPN) to pick P/S phases from continuous seismic recordings. It first classifies the phase type for a segment of waveform, and then performs regression to get accurate phase arrival time. The network is optimized using deep separable convolution to reduce the number of trainable parameters and improve its computation efficiency. Experiments using the STanford EArthquake Dataset (STEAD) show that the precision of LPPN can reach 95.2% and 83.7% with the recalls 94.4% and 84.7% for P and S phases, respectively. The classification–regression approach shows comparable performance to traditional point‐to‐point methods with lower computation cost. LPPN can be configured to have different model size and run on a wide range of devices. It is open‐source and can support phase picking for large‐scale dataset or in other speed sensitive scenarios.

Vehicle-mounted ground-penetrating radar (GPR) has been used to non-destructively inspect and evaluate railway subgrade conditions. However, existing GPR data processing and interpretation methods mostly rely on time-consuming manual interpretation, and limited studies have applied machine learning methods. GPR data are complex, high-dimensional, and redundant, in particular with non-negligible noises, for which traditional machine learning methods are not effective when applied to GPR data processing and interpretation. To solve this problem, deep learning is more suitable to process large amounts of training data, as well as to perform better data interpretation. In this study, we proposed a novel deep learning method to process GPR data, the CRNN network, which combines convolutional neural networks (CNN) and recurrent neural networks (RNN). The CNN processes raw GPR waveform data from signal channels, and the RNN processes features from multiple channels. The results show that the CRNN network achieves a higher precision at 83.4%, with a recall of 77.3%. Compared to the traditional machine learning method, the CRNN is 5.2 times faster and has a smaller size of 2.6 MB (traditional machine learning method: 104.0 MB). Our research output has demonstrated that the developed deep learning method improves the efficiency and accuracy of railway subgrade condition evaluation.

Joint inversion of multitype datasets is an effective approach for high-precision subsurface imaging. We present a new deep learning-based method to jointly invert Rayleigh wave phase velocity and ellipticity into shear-wave velocity of the crust and uppermost mantle. A multimodal deep neural network (termed JointNet) is designed to analyze these two independent physical parameters and generate outputs, including velocity and layer thicknesses. JointNet is trained using random 1D models and corresponding synthetic phase velocity and ellipticity, resulting in a low cost for the training dataset. Evaluation using synthetic and observed data shows that JointNet produces highly comparable results compared to those from a Markov chain Monte Carlo-based method and significantly improves inversion speed. Training using synthetic data ensures its generalized application in various regions with different velocity structures. Moreover, JointNet can be easily extended to include additional datatypes and act as a joint inversion framework to further improve imaging resolution.

SUMMARY: We present an earthquake catalogue in northeastern Tibetan plateau between September 2013 to April 2016 during the ChinArray-II deployment. Using continuous records from 676 transportable ChinArray-II stations and 172 permanent stations, the P/S phases are obtained using one deep learning phase picker. After associating these phases, the events are identified and located to establish the ChinArray-II Regional Earthquake Database (CARED-II). Benefiting from both improved station coverage and sensitive phase picker, CARED-II catalogue has 156 057 events (around 3 million picks), about tenfold more than the manual routine catalogue (15 967 events) using the permanent stations. The improved event catalogue delineates the fault structures clearly. The deep structure of south-dipping north Qilian thrust faults is revealed, consisting with previous geology studies. The hidden faults and fault connectivity are revealed by improved seismicity, especially in the Alxa Block with sparse permanent stations and severe environments restricting geological field work. Moreover, small anthropogenic events are identified and related to highway tunnel construction across Qinling Mountain, forming a straight event cluster. The results demonstrate the high event detection ability of our procedure and reliability of the automatic catalogue. Our array-based CARED-II catalogue provides improved seismicity images in northeastern Tibet and could be used for further seismology and geotectonic studies.

We present SeismicXM, a cross-task foundation model designed for single-station seismic waveform analysis. Unlike conventional approaches that target individual tasks, SeismicXM simultaneously performs seismic phase picking (Pg, Sg, Pn, and Sn), P-wave first-motion polarity classification, and event-type classification directly from raw three-component waveforms. The architecture integrates a convolutional encoder with a bidirectional Transformer backbone and multiple task-specific decoders, enabling shared representation learning while flexibly producing task-dependent outputs. SeismicXM is pretrained on the large-scale Comprehensive Dataset of Chinese Seismic Network dataset, which contains more than 45 million annotated phases, polarities, and event labels. Leveraging this diverse dataset, the model learns generalizable seismic representations that can be efficiently adapted to new regions with minimal labeled data. Benchmark experiments demonstrate strong performance: SeismicXM achieves F1 scores above 0.88 for Pg and Sg picking, and reaches F1 scores of 0.871 and 0.687 for Pn and Sn under full-waveform evaluation. The model also attains over 0.97 area under the curve in P-wave polarity classification and approximately 94% accuracy in event-type classification after fine-tuning on a small regional dataset. These results indicate that SeismicXM provides an effective and transferable solution for diverse waveform-processing tasks, highlighting the potential of pretrained multitask Transformers to streamline operational earthquake monitoring and analysis workflows.

The effective utilization of medium-high temperature geothermal energy is pivotal in reducing carbon emissions and plays a crucial role in developing clean energy technologies. The MiDu geothermal field, situated in the southeastern region of Dali Prefecture, Yunnan Province, lies within the Mediterranean–Himalayan high-temperature geothermal belt and is characterized by abundant geothermal resources. However, due to its considerable depth, exploration poses significant risks, resulting in a total utilization rate of less than 0.5% of the total reserves. This study employs natural seismic data to perform a tomographic analysis of the geothermal system in the Midu basin. By examining the P-wave velocity (Vp) and the velocity ratio of P-waves and S-waves (Vp/Vs) at various depths, the findings reveal that the basin comprises two distinct structural layers: the thrust basement of the Mesozoic and Paleozoic eras and the strike–slip extensional sedimentary layer of the Cenozoic era. A low-velocity anomaly in the central basin corresponds to the loose Cenozoic sedimentary layer. In contrast, high-velocity anomalies at the basin edges correlate with boundary faults and the Mesozoic–Paleozoic strata. Below a depth of 4 km, the Red River Fault and MiDu Fault continue to dominate the basin’s structure, whereas the influence of the Malipo Fault diminishes. The MiDu Fault exhibits higher thermal conductivity than the Yinjie Fault. It interfaces with multiple carbonate and basalt formations characterized by well-developed pores and fractures, making it a crucial conduit for water and a control point for geothermal storage. Consequently, the existence of medium-high temperature (>90 °C) geothermal resources for power generation should be concentrated around the Midu fault on the western side of the basin, while the Yinjie fault area is more favorable for advancements in heating and wellness.