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.

Untuned large‐volume airgun arrays are now widely used in onshore lakes and rivers to image crustal structure and monitor its temporal variation. However, the property of the airgun source is not well understood. In this article, we investigate the source signature of an untuned airgun array by comparing seismic records from closely located airgun shots, a chemical explosion, and a small earthquake. We find that the airgun signal has a dominant frequency of 2–8 Hz, which is the lowest among the three sources. Because of its relative low frequency, the seismic wave can propagate up to $∼40km$, which is longer than that for an earthquake with similar magnitude ($ML$ 0.7). After being stacked for 20 times or more, the airgun signal can be detected as far as 150 km. The travel times of different P‐wave phases from the airgun source match that picked from the explosion profile. However, the airgun waveforms are slightly more complicated, likely due to reverberations of the air bubbles. However, deconvolutions of near‐source recording only result in minor improvement in the sharpness of phase arrivals. In addition, we show that the signal‐to‐noise ratios of the stacked trace do not depend strongly on the stacking numbers, making it possible to achieve a high temporal resolution on the order of hours. Although the results in this article are drawn from a case study, they have implications for the use of airgun arrays in onshore exploration in other regions.

The development of sensitive automatic phase picking method can produce P/S phase picks for lots of seismic stations quickly. This raises the requirement for fast phase association method to group phase picks into candidate events. Here we present FastLink, a machine learning and GPU based phase associator, to group phase picks efficiently. FastLink is designed to utilize parallel computing power of GPU, it uses a 3-layer neural network to determine valid picks and the non-maximum suppression (NMS) algorithm to distinguish the events simultaneously occurred in different places. With the aid of GPU, FastLink can perform faster grid search than traditional methods. We apply FastLink to aftershock sequences identification of the Yangbi M_s 6.4 earthquake occurred on 21 May 2021, in Yunnan Province, China. Over 17 000 events are successfully associated using about 2 million picks from 1-week records of 79 stations surrounding the main shock. Over 93.5 per cent of the M_L ≥ 0 events in the manual verified bulletin M_L are successfully associated by FastLink. Compared with the widely used REAL method, FastLink produces comparable results and is over 12 times faster with GPU acceleration. This shows its potential for processing a large number of phase picks quickly.