| 27 Mar 2026
A Study of Pre-earthquake Anomalies in Geomagnetic and Borehole Strain Data Based on the Pyraformer Model
Abstract. As a major natural hazard, earthquakes are accompanied by lithospheric stress accumulation and anomalies in multiple physical fields during their preparation process. Identifying pre-earthquake anomalies is therefore of great scientific significance for understanding earthquake preparation and improving earthquake prediction. Taking the 2013 Lushan earthquake in Sichuan, China (Ms 7.0), as a case study, this work presents a comprehensive analysis of geomagnetic and borehole strain observations to investigate the spatiotemporal evolution of pre-earthquake multi-physical-field anomalies. First, Variational Mode Decomposition (VMD) was applied to geomagnetic and borehole strain data to achieve multi-scale signal decomposition, thereby separating long-term background variations from short-period disturbance components. Subsequently, the Pyraformer model was employed to model and predict the high-frequency feature sequences, and the prediction residuals were extracted to quantify pre-earthquake anomalies. The results show that borehole strain anomalies first emerged along local fault segments during the pre-seismic stage and gradually accumulated, whereas geomagnetic anomalies subsequently intensified and accelerated at adjacent stations. Overall, the anomalies exhibited a spatiotemporal evolution pattern characterized by local-to-regional expansion and stage-wise accelerated accumulation. Further analysis indicates that borehole strain anomalies generally preceded geomagnetic anomalies, with a time lag ranging from several days to several weeks between the rapid accumulation stage of borehole strain anomalies and the strengthening stage of geomagnetic anomalies. In addition, the cumulative anomaly curves commonly displayed a pronounced S-shaped growth pattern, reflecting the gradual increase in stress-strain during earthquake preparation and its subsequent transfer to the electromagnetic field response. These findings reveal a coupling relationship between the stress-strain field and the electromagnetic field during the earthquake preparation process. The integrated VMD-Pyraformer framework proposed in this study demonstrates strong stability and reliability in identifying geomagnetic and borehole strain anomalies and provides new observational evidence and an effective analytical approach for multi-station, multi-parameter joint monitoring and pre-earthquake anomaly identification.
The manuscript entitled ‘‘A Study of Pre-earthquake Anomalies in Geomagnetic and Borehole Strain Data Based on the Pyraformer Model’‘ presents the results of analyzing magnetic and strain data for a large magnitude earthquake aiming to identify pre-seismic signal using Deep Learning techniques. The researchers used advanced computer techniques to separate normal background changes from unusual signals that might happen before the 2013 Lushan earthquake in Sichuan, China.
My general feeling reading the manuscript is that is well organized, interesting work providing some new useful information of exploiting multi parametric data to estimate/define pre-seismic signal on geomagnetic and strain time series. I think that the manuscript requires some small modifications in order to be improved.  In any case, it is considered, in my opinion, as a nice work well organized, with a full ‘‘background’‘ information and interesting results.
A general comment that I have is that the text and the processing focus on the pre-seismic period and ignores completely the post seismic period. Of course, it is well explained by the authors, that the manuscript examines the possible anomalies on the preparatory stage of a strong earthquake, but since the earthquake occurred in 2013, and a long time passed since this event, it would be interesting to see if the area, and the geomagnetic and strain data from the same stations,  shows any similar behavior as during the pre-seismic period. In other words, as a validation of this processing it could be interesting to examine the years after the earthquake if there was another period that these magnetic and strain stations show anomalous days without the occurrence of an earthquake, BUT still this may be the subject for another publication.
The authors used geomagnetic data only few months before the earthquake (since January 2013). In lines 299-301 :’‘…were selected to characterise the short- to medium-term background variations…’‘  the authors clearly describe that they focus on  ‘‘short- to medium-term’‘ on geomagnetic data, but the strain data cover longer period … >1 year ‘‘….to characterise the strain evolution over a longer time scale’‘. I think that they just have to add a more well justified explanation for this discrepancy concerning the time span of the two data sets.
Some more detail comments:
Â
Finally, Lines 528-534 – ‘‘ Conclusion ‘‘ . I think that the authors have to rewrite this section.
It is a very short text to close such a nice work! Â
The authors have to provide a bit more specific statements about their results and finding and not just write general ‘‘declarations’‘ like: ‘‘These results indicate that borehole strain may capture the earlier response of local stress accumulation, whereas geomagnetic anomalies may reflect the subsequent electromagnetic response.’‘
So my suggestion is that the ‘‘Conclusion’‘ section has to specifically address the methodologies, findings and results of this manuscript in order the reader to be able to clearly understand what this research provides.