Albert Leonardo Aguilar Suarez

arXiv2026

Global Prevalence of Anti-similar Earthquakes at Intermediate Depth

Anti-similar earthquakes, those with nearly identical but flipped waveforms, remain an uncommon observation. We document three locations where these events occur at intermediate depth. The first of these is Colombia, including the well-established source of anti-similar events in the Bucaramanga Nest, and new reports in other intermediate depth seismicity clusters in the Wadati-Benioff zones of Colombia. The second is the Alaska subduction zone, and the third is the Vrancea seismic nest in Romania. These new observations, combined with reports from Japan and Chile, provide mounting evidence that anti-similar earthquake sequences might be a common or characteristic process of intermediate-depth seismicity. Either conjugate faulting promoted by eclogitization induced stresses or sub-parallel faulting can explain their occurrence at these depths. Careful examination of seismicity in other productive intermediate depth seismicity zones might reveal an even more widespread geography for anti-similar earthquakes and place tighter constraints on their underlying mechanism.

physics.geo-ph↗

arXiv2025

Pervasive Label Errors in Seismological Machine Learning Datasets

The recent boom in artificial intelligence and machine learning has been powered by large datasets with accurate labels, combined with algorithmic advances and efficient computing. The quality of data can be a major factor in determining model performance. Here, we detail observations of commonly occurring errors in popular seismological machine learning datasets. We used an ensemble of available deep learning models PhaseNet and EQTransformer to evaluate the dataset labels and found four types of errors ranked from most prevalent to least prevalent: (1) unlabeled earthquakes; (2) noise samples that contain earthquakes; (3) inaccurately labeled arrival times, and (4) absent earthquake signals. We checked a total of 8.6 million examples from the following datasets: Iquique, ETHZ, PNW, TXED, STEAD, INSTANCE, AQ2009, and CEED. The average error rate across all datasets is 3.9 %, ranging from nearly zero to 8 % for individual datasets. These faulty data and labels are likely to degrade model training and performance. By flagging these errors, we aim to increase the quality of the data used to train machine learning models, especially for the measurement of arrival times, and thereby to improve the reliability of the models. We present a companion list of examples that contain problems, aiming to integrate them into training routines so that only the reliable data is used for training.

physics.geo-ph↗