Forecasting future earthquakes with deep neural networks: application to California

被引:0
作者
Zhang, Ying [1 ,2 ,3 ]
Zhan, Chengxiang [4 ]
Huang, Qinghua [2 ]
Sornette, Didier [5 ]
机构
[1] Univ Sci & Technol Beijing, Sch Automat & Elect Engn, Beijing 100083, Peoples R China
[2] Peking Univ, Dept Geophys, Sch Earth & Space Sci, Beijing 100871, Peoples R China
[3] Beijing Engn Res Ctr Ind Spectrum Imaging, Beijing 100083, Peoples R China
[4] China Univ Geosci Beijing, Sch Sci, Beijing 100083, Peoples R China
[5] Southern Univ Sci & Technol SUSTech, Inst Risk Anal Predict & Management Risks X, Shenzhen 518055, Peoples R China
基金
中国国家自然科学基金;
关键词
Machine learning; Probabilistic forecasting; Earthquake interaction; forecasting and prediction; Statistical seismology; SHORT-TERM; SEISMICITY; MODELS; STRATEGIES; PREDICTION; STATE; LAW;
D O I
10.1093/gji/ggae373
中图分类号
P3 [地球物理学]; P59 [地球化学];
学科分类号
0708 ; 070902 ;
摘要
We use the spatial map of the logarithm of past estimated released earthquake energies as input of fully convolutional networks (FCN) to forecast future earthquakes. This model is applied to California and compared with an elaborated version of the epidemic type aftershock sequence (ETAS) model. Our long-term earthquake forecast simulations show that the FCN model is close to the ETAS model in forecasting earthquakes with M >= 3.0 , 4.0 , and 5.0 according to the Molchan diagram. Moreover, training and implementing the FCN model is 2000-4000 times faster than calibrating the ETAS model and generating its probabilistic forecasts. The FCN model is straightforward in terms of its neural network structure and feature engineering. It does not require extensive knowledge of statistical seismology or the analysis of earthquake catalogue completeness. Using the earthquake catalogue with M >= 0 as FCN input can enhance the model's performance in some time-magnitude forecasting windows.
引用
收藏
页码:81 / 95
页数:15
相关论文
共 59 条
[11]  
Gutenberg B., 1944, Bulletin of the Seismological Society of America, V34, P185, DOI [DOI 10.1785/BSSA0340040185, 10.1785/BSSA0340040185]
[12]  
HAWKES AG, 1971, J ROY STAT SOC B, V33, P438
[13]   CLUSTER PROCESS REPRESENTATION OF A SELF-EXCITING PROCESS [J].
HAWKES, AG ;
OAKES, D .
JOURNAL OF APPLIED PROBABILITY, 1974, 11 (03) :493-503
[14]   Comparison of short-term and time-independent earthquake forecast models for southern California [J].
Helmstetter, A ;
Kagan, YY ;
Jackson, DD .
BULLETIN OF THE SEISMOLOGICAL SOCIETY OF AMERICA, 2006, 96 (01) :90-106
[15]   Foreshocks explained by cascades of triggered seismicity [J].
Helmstetter, A ;
Sornette, D .
JOURNAL OF GEOPHYSICAL RESEARCH-SOLID EARTH, 2003, 108 (B10)
[16]   Mainshocks are aftershocks of conditional foreshocks: How do foreshock statistical properties emerge from aftershock laws [J].
Helmstetter, A ;
Sornette, D ;
Grasso, JR .
JOURNAL OF GEOPHYSICAL RESEARCH-SOLID EARTH, 2003, 108 (B1)
[17]   Precursory seismicity changes associated with the Mw=7.4 1999 August 17 Izmit (Turkey) earthquake [J].
Huang, Q ;
Öncel, AO ;
Sobolev, GA .
GEOPHYSICAL JOURNAL INTERNATIONAL, 2002, 151 (01) :235-242
[18]   Search for reliable precursors: A case study of the seismic quiescence of the 2000 western Tottori prefecture earthquake [J].
Huang, QH .
JOURNAL OF GEOPHYSICAL RESEARCH-SOLID EARTH, 2006, 111 (B4)
[19]   Seismic quiescence before the 2000 M=7.3 Tottori earthquake -: art. no. 1578 [J].
Huang, QH ;
Nagao, T .
GEOPHYSICAL RESEARCH LETTERS, 2002, 29 (12) :19-1
[20]   Forecasting the epicenter of a future major earthquake [J].
Huang, Qinghua .
PROCEEDINGS OF THE NATIONAL ACADEMY OF SCIENCES OF THE UNITED STATES OF AMERICA, 2015, 112 (04) :944-945