Abstract
Aftershocks can cause disasters again after mainshocks, which result in threat to life and economic loss. In order to avoid secondary disasters, it is necessary to predict whether aftershocks would happen in a given region. There have been studies using different features and methods to predict aftershocks spatial distribution. However, it is still unclear which are more important for aftershock prediction, input features or models; which type of features is more predictive for the prediction task. In this paper, we predict aftershock spatial distribution by combining different types of features and applying different machine learning methods. We introduce five different types of features and combine them together for prediction: the stress change sensors, their logarithmic values, the physical quantities, the magnitude of mainshocks, and the distance between the grid cell and the epicenter of mainshocks. We train different classifiers: Naive Bayes, Support Vector Machine,Gradient Boosting Decision Tree, k-Nearest Neighbors, Logistic regression, and DMAP (a Deep Neural Network model). Based on the 62,811 aftershocks of 171 distinct mainshocks in the past about 40 years in China, we conduct comprehensive experiments and analyses. We find that features play a more important role for this prediction task. Using the same feature type, different classifiers obtain quite similar performance. With different features, the same model performs differently. Taking the combined features as input, we achieve the state-of-the-art performance, with an AUC of 0.9530, about 4% higher than that of DeVries et al., showing the superiority of the combined features. Among all the features, adding the distance to the stress change sensors contributes the most to improve the prediction performance. In addition, it is found that the model prediction performance varies in terms of the time spans after mainshocks and the aftershock magnitudes.
Similar content being viewed by others
Availability of data and material
All the experiments related to machine learning methods were implemented by Python. The code and the used dataset are available on GitHub(https://github.com/whyboy/aftershockprediction).
Code availability
The code and the used dataset are available on GitHub(https://github.com/whyboy/aftershockprediction).
References
Akram J, Ovcharenko O, Peter D (2017) A robust neural network-based approach for microseismic event detection
Båth M (1965) Lateral inhomogeneities of the upper mantle. Tectonophysics 2 (6):483–514
Burges CJ (1998) A tutorial on support vector machines for pattern recognition. Data mining and knowledge discovery 2(2):121–167
Carlson JM, Langer JS, Shaw BE (1994) Dynamics of earthquake faults. Rev Mod Phys 66(2):657
Coban KH, Sayil N (2019) Evaluation of earthquake recurrences with different distribution models in western anatolia. J Seismol 23(6):1405–1422
DeVries PM, Viégas F, Wattenberg M, Meade BJ (2018) Deep learning of aftershock patterns following large earthquakes. Nature 560(7720):632–634
Fukunaga K, Narendra PM (1975) A branch and bound algorithm for computing k-nearest neighbors. IEEE transactions on computers 100(7):750–753
Geist EL, Parsons T (2018) Determining on-fault earthquake magnitude distributions from integer programming. Computers & Geosciences 111:244–259
Gutenberg B, Richter CF (1944) Frequency of earthquakes in california. Bull Seismol Soc Am 34(4):185–188
Hand DJ, Till RJ (2001) A simple generalisation of the area under the roc curve for multiple class classification problems. Machine learning 45(2):171–186
Jacques E, King G, Tapponnier P, Ruegg J, Manighetti I (1996) Seismic activity triggered by stress changes after the 1978 events in the Asal Rift, Djibouti. Geophysical research letters 23(18):2481–2484
Kappler KN, Schneider DD, MacLean LS, Bleier T, Lemon J (2019) An algorithmic framework for investigating the temporal relationship of magnetic field pulses and earthquakes applied to california. Computers & Geosciences 133:104–317
Karimzadeh S, Matsuoka M, Kuang J, Ge L (2019) Spatial prediction of aftershocks triggered by a major earthquake: a binary machine learning perspective. ISPRS International Journal of Geo-Information 8(10):462
Ke G, Meng Q, Finley T, Wang T, Chen W, Ma W, Ye Q, Liu TY (2017) Lightgbm: a highly efficient gradient boosting decision tree. In: Advances in neural information processing systems, pp 3146–3154
King GC, Stein RS, Lin J (1994) Static stress changes and the triggering of earthquakes. Bull Seismol Soc Am 84(3):935–953
Kleinbaum DG, Dietz K, Gail M, Klein M, Klein M (2002) Logistic regression. Springer
LeCun Y, Bengio Y, Hinton G (2015) Deep learning. Nature 521(7553):436–444
Lewis DD (1998) Naive (bayes) at forty: the independence assumption in information retrieval. In: European conference on machine learning, Springer, pp 4–15
Li A, Kang L (2009) Knn-based modeling and its application in aftershock prediction. In: 2009 International Asia Symposium on Intelligent Interaction and Affective Computing, IEEE, pp 83–86
Li Z, Meier MA, Hauksson E, Zhan Z, Andrews J (2018) Machine learning seismic wave discrimination: application to earthquake early warning. Geophys Res Lett 45(10):4773–4779
Meade BJ, DeVries PM, Faller J, Viegas F, Wattenberg M (2017) What is better than coulomb failure stress? a ranking of scalar static stress triggering mechanisms from 105 mainshock-aftershock pairs. Geophys Res Lett 44(22):11–409
Meier MA, Werner M, Woessner J, Wiemer S (2014) A search for evidence of secondary static stress triggering during the 1992 mw7. 3 landers, california, earthquake sequence. Journal of Geophysical Research:, Solid Earth 119(4):3354–3370
Mignan A (2019) A preliminary text classification of the precursory accelerating seismicity corpus: inference on some theoretical trends in earthquake predictability research from 1988 to 2018. J Seismol 23(4):771–785
Mignan A, Broccardo M (2019) One neuron versus deep learning in aftershock prediction. Nature 574(7776):E1–E3
Mignan A, Broccardo M (2020a) Comment on “elastic strain energy and pore-fluid pressure control of aftershocks” by terakawa et al.[earth planet. sci. lett. 535 (2020) 116103]. Earth and Planetary Science Letters 544:116,402
Mignan A, Broccardo M (2020b) Neural network applications in earthquake prediction (1994–2019): meta-analytic and statistical insights on their limitations. Seismological Research Letters
Nostro C, Cocco M, Belardinelli ME (1997) Static stress changes in extensional regimes: an application to southern apennines (italy). Bull Seismol Soc Am 87(1):234–248
Nur A, Mavko G (1974) Postseismic viscoelastic rebound. Science 183(4121):204–206
Ochoa LH, Niño LF, Vargas CA (2018) Fast magnitude determination using a single seismological station record implementing machine learning techniques. Geodesy and Geodynamics 9(1):34–41
Okada Y (1992) Internal deformation due to shear and tensile faults in a half-space. Bulletin of the seismological society of America 82(2):1018–1040
Omori F (1894) On the after-shocks of earthquakes, vol 7. The University
Parsons T (2020) On the use of receiver operating characteristic tests for evaluating spatial earthquake forecasts. Geophysical Research Letters 47(17):e2020GL088,570
Parsons T, Stein RS, Simpson RW, Reasenberg PA (1999) Stress sensitivity of fault seismicity: a comparison between limited-offset oblique and major strike-slip faults. Journal of Geophysical Research: Solid Earth 104(B9):20,183–20,202
Perol T, Gharbi M, Denolle M (2018) Convolutional neural network for earthquake detection and location. Science Advances 4(2):e1700,578
Reasenberg PA, Simpson RW (1992) Response of regional seismicity to the static stress change produced by the loma prieta earthquake. Science 255(5052):1687–1690
Robinson R, Zhou S (2005) Stress interactions within the tangshan, china, earthquake sequence of 1976. Bull Seismol Soc Am 95(6):2501–2505
Ross ZE, Meier MA, Hauksson E, Heaton TH (2018) Generalized seismic phase detection with deep learning. Bull Seismol Soc Am 108(5A):2894–2901
Rydelek P, Sacks I (1990) Asthenospheric viscosity and stress diffusion: a mechanism to explain correlated earthquakes and surface deformations in ne japan. Geophys J Int 100(1):39–58
Scitovski S (2018) A density-based clustering algorithm for earthquake zoning. Computers & Geosciences 110:90–95
Sharma S, Hainzl S, Zöeller G, Holschneider M (2020) Is coulomb stress the best choice for aftershock forecasting? Journal of Geophysical Research: Solid Earth 125(9):e2020JB019, 553
Spence W, Sipkin SA, Choy GL (1989) Measuring the size of an earthquake. Earthquake Information Bulletin (USGS) 21(1):58–63
Terakawa T, Matsu’ura M, Noda A (2020) Elastic strain energy and pore-fluid pressure control of aftershocks. Earth and Planetary Science Letters 535:116,103
Toda S, Stein RS, Reasenberg PA, Dieterich JH, Yoshida A (1998) Stress transferred by the 1995 mw= 6.9 kobe, japan, shock: Effect on aftershocks and future earthquake probabilities. Journal of Geophysical Research:, Solid Earth 103(B10):24,543–24,565
Troise C, De Natale G, Pingue F, Petrazzuoli S (1998) Evidence for static stress interaction among earthquakes in the south–central apennines (italy). Geophys J Int 134(3):809–817
Utsu T (1961) A statistical study on the occurrence of aftershocks. Geophys Mag 30:521–605
Wang M, Shen J, Pan Z, Han D (2019) An improved supported vector regression algorithm with application to predict aftershocks. J Seismol 23(5):983–993
Zhou Y, Yue H, Kong Q, Zhou S (2019) Hybrid event detection and phase-picking algorithm using convolutional and recurrent neural networks. Seismol Res Lett 90(3):1079–1087
Zhu W, Beroza GC (2019) Phasenet: a deep-neural-network-based seismic arrival-time picking method. Geophys J Int 216(1):261–273
Funding
This document is the result of the research project funded by the National Key R&D Program of China under Grant (No. 2018YFC1504006), and National Natural Science Foundation of China (No. 61802342, 61802340).
Author information
Authors and Affiliations
Contributions
Sha Zhao wrote the paper and designed the experiments; Haiyan Wang did the data pre-processing and conducted the experiments related to machine learning methods; Yan Xue did the data acquisition and conducted the experiments related to physical methods; Shijian Li and Yilin Wang analyzed the data; Jie Liu carried out the acquisition; Gang Pan designed the experiments and analyzed the data. All the authors have read and revised the manuscript. Dr Xue and Dr Pan are the corresponding authors.
Corresponding author
Ethics declarations
Conflict of interest
The authors declare no competing interests.
Additional information
Publisher’s note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Rights and permissions
About this article
Cite this article
Zhao, S., Wang, H., Xue, Y. et al. What are more important for aftershock spatial distribution prediction, features, or models? A case study in China. J Seismol 26, 181–196 (2022). https://doi.org/10.1007/s10950-021-10044-x
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s10950-021-10044-x