Interpretability and spatial efficacy of a deep-learning-based on-site early warning framework using explainable artificial intelligence and geographically weighted random forests

Abstract

Earthquakes pose significant risks globally, necessitating effective seismic risk mitigation strategies like earthquake early warning (EEW) systems. However, developing and optimizing such systems requires thoroughly understanding their internal procedures and coverage limitations. This study examines a deep-learning-based on-site EEW framework known as ROSERS (Real-time On-Site Estimation of Response Spectra) proposed by the authors, which constructs response spectra from early recorded ground motion waveforms at a target site. This study has three primary goals: (1) evaluating the effectiveness and applicability of ROSERS to subduction seismic sources; (2) providing a detailed interpretation of the trained deep neural network (DNN) and surrogate latent variables (LVs) implemented in ROSERS; and (3) analyzing the spatial efficacy of the framework to assess the coverage area of on-site EEW stations. ROSERS is retrained and tested on a dataset of around 11,000 unprocessed Japanese subduction ground motions. Goodness-of-fit testing shows that the ROSERS framework achieves good performance on this database, especially given the peculiarities of the subduction seismic environment. The trained DNN and LVs are then interpreted using game theory-based Shapley additive explanations to establish cause-effect relationships. Finally, the study explores the coverage area of ROSERS by training a novel spatial regression model that estimates the LVs using geographically weighted random forest and determining the radius of similarity. The results indicate that on-site predictions can be considered reliable within a 2–9 km radius, varying based on the magnitude and distance from the earthquake source. This information can assist end-users in strategically placing sensors, minimizing blind spots, and reducing errors from regional extrapolation.

Keywords

Earthquake early warning systems / Spatial regression / Neural networks / Japanese subduction / Explainable artificial intelligence

CRediT authorship contribution statement

Jawad Fayaz: Conceptualization, Data curation, Investigation, Methodology, Writing – original draft. Carmine Galasso: Investigation, Validation, Writing – review & editing.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Acknowledgements

The authors extend their appreciation to Mr. Owais Javed, graduate student at Teesside University, for his contributions to the organization of data and his diligent efforts in the preliminary trials of spatial regression modeling. The authors also acknowledge the assistance of Dr. Rodrigo Astroza and his group in compiling the Japanese subduction ground-motion database.

References

[1] Akazawa, T., 2004. A technique for automatic detection of onset time of P- and S-phases in strong motion records, in: Proceed of the 13th World Conference on Earthquake Engineering, Vancouver, B.C., Canada, August 1-6, Paper No. 786.

[2] T.D. Ancheta, R.B. Darragh, J.P. Stewart, E. Seyhan, W.J. Silva, B.-S.-J. Chiou, K.E. Wooddell, R.W. Graves, A.R. Kottke, D.M. Boore, T. Kishida, J.L. Donahue. NGA-West2 database. Earthq. Spectra, 30 (3) (2014), pp. 989-1005,

[3] Aydınoğlu, M.N., Vuran, E., 2015. Developments in Seismic Design of Tall Buildings: Preliminary Design of Coupled Core Wall Systems. In: Ansal, A. (Ed.), Perspectives on European Earthquake Engineering and Seismology. Geotechnical, Geological and Earthquake Engineering, vol 39. Springer, Cham, 227–243. https://doi.org/10.1007/978-3-319-16964-4_9.

[4] Bergstra, J., Bengio, Y., 2012. Random search for hyper-parameter optimization. J. Mach. Learn. Res. 13, 281–305. http://scikit-learn.sourceforge.net.

[5] R. Bhardwaj, M.L. Sharma, A. Kumar. Multi-parameter algorithm for earthquake early warning. Geomat. Nat. Haz. Risk, 7 (4) (2016), pp. 1242-1264,

[6] K. Bi, H. Hao. Modelling and simulation of spatially varying earthquake ground motions at sites with varying conditions. Probab. Eng. Mech., 29 (2012), pp. 92-104,

[7] S. Bloemheuvel, J. van den Hoogen, D. Jozinović, A. Michelini, M. Atzmueller. Graph neural networks for multivariate time series regression with application to seismic data. Int. J. Data Sci. Anal., 16 (3) (2023), pp. 317-332,

[8] D. Bosq. Sufficiency and efficiency in statistical prediction. Statist. Probab. Lett., 77 (3) (2007), pp. 280-287,

[9] Bozorgnia, Y., Vitelmo V. Bertero, V.V., 2004. Earthquake Engineering: From Engineering Seismology to Performance-Based Engineering. CRC Press, Boca Raton. https://doi.org/10.1201/9780203486245.

[10] Campbell, K.W., Bozorgnia, Y., 2013. NGA-West2 Campbell-Bozorgnia ground motion model for the horizontal components of PGA, PGV, response spectra for periods ranging from 0.01 to 10 Sec. PEER Report 2013/06. Peer Report, no. May.

[11] K.W. Campbell, Y. Bozorgnia. Ground motion models for the horizontal components of Arias Intensity (AI) and cumulative absolute velocity (CAV) using the NGA-West2 database. Earthq. Spectra, 35 (3) (2019), pp. 1289-1310,

[12] L. Caramenti, A. Menafoglio, S. Sgobba, G. Lanzano. Multi-source geographically weighted regression for regionalized ground-motion models. Spatial Statistics, 47 (2022), Article 100610,

[13] A. Caruso, S. Colombelli, L. Elia, M. Picozzi, A. Zollo. An on-site alert level early warning system for Italy. J. Geophys. Res. Solid Earth, 122 (3) (2017), pp. 2106-2118,

[14] G. Cremen, C. Galasso. Earthquake early warning: recent advances and perspectives. Earth Sci. Rev., 205 (2020), Article 103184,

[15] M. Ester, H.-P. Kriegel, J. Sander, X. Xu. A density-based algorithm for discovering clusters in large spatial databases with noise. Knowl. Discov. Data Min., 96 (34) (1996), pp. 226-231

[16] J. Fayaz, C. Galasso. A deep neural network framework for real-time on-site estimation of acceleration response spectra of seismic ground motions. Comput. Aided Civ. Inf. Eng., 38 (1) (2022), pp. 87-103,

[17] J. Fayaz, M. Medalla, F. Zareian. Sensitivity of the response of Box-Girder Seat-type bridges to the duration of ground motions arising from crustal and subduction earthquakes. Eng. Struct., 219 (2020), Article 110845,

[18] J. Fayaz, Y. Xiang, F. Zareian. Generalized ground motion prediction model using hybrid recurrent neural network. Earthq. Eng. Struct. Dyn., 50 (6) (2021), pp. 1539-1561,

[19] J. Fayaz, M. Medalla, P. Torres-Rodas, C. Galasso. A recurrent-neural-network-based generalized ground-motion model for the Chilean subduction seismic environment. Struct. Saf., 100 (2023), Article 102282,

[20] E.H. Field, T.E. Dawson, K.R. Felzer, A.D. Frankel, V. Gupta, T.H. Jordan, T. Parsons, M.D. Petersen, R.S. Stein, R.J. Weldon, C.J. Wills. Uniform California earthquake rupture forecast, version 2 (UCERF 2). Bull. Seismol. Soc. Am., 99 (4) (2009), pp. 2053-2107,

[21] E.H. Field, T.H. Jordan, M.T. Page, K.R. Milner, B.E. Shaw, T.E. Dawson, G.P. Biasi, T. Parsons, et al.. A synoptic view of the third uniform California earthquake rupture forecast (UCERF3). Seismol. Res. Lett., 88 (5) (2017), pp. 1259-1267,

[22] C. Galasso, E. Zuccolo, K. Aljawhari, G. Cremen, N.S. Melis. Assessing the potential implementation of earthquake early warning for schools in the Patras region, Greece. Int. J. Disaster Risk Reduct., 90 (2023), Article 103610,

[23] S. Georganos, T. Grippa, A. Niang Gadiaga, C. Linard, M. Lennert, S. Vanhuysse, N. Mboga, E. Wolff, S. Kalogirou. Geographical random forests: a spatial extension of the random forest algorithm to address spatial heterogeneity in remote sensing and population modelling. Geocarto Int., 36 (2) (2021), pp. 121-136,

[24] T.Y. Hsu, S.K. Huang, Y.W. Chang, C.H. Kuo, C.M. Lin, T.M. Chang, K.L. Wen, C.H. Loh. Rapid on-site peak ground acceleration estimation based on support vector regression and P-wave features in Taiwan. Soil Dyn. Earthq. Eng., 49 (2013), pp. 210-217,

[25] A.G. Iaccarino, M. Picozzi, D. Bindi, D. Spallarossa. Onsite earthquake early warning: predictive models for acceleration response spectra considering site effects. Bull. Seismol. Soc. Am., 110 (3) (2020), pp. 1289-1304,

[26] D. Jozinović, A. Lomax, I. Štajduhar, A. Michelini. Rapid prediction of earthquake ground shaking intensity using raw waveform data and a convolutional neural network. Geophys. J. Int., 222 (2) (2020), pp. 1379-1389,

[27] D. Jozinović, A. Lomax, I. Štajduhar, A. Michelini. Transfer learning: Improving neural network based prediction of earthquake ground shaking for an area with insufficient training data. Geophys. J. Int., 229 (1) (2022), pp. 704-718,

[28] E. Kalkan. An automatic P-phase arrival-time picker. Bull. Seismol. Soc. Am., 106 (3) (2016), pp. 971-986,

[29] D.P. Kingma, M. Welling. An introduction to variational autoencoders. Found. Trends Mach. Learn., 12 (4) (2019), pp. 307-392,

[30] S.L. Kramer. Geotechnical Earthquake Engineering. Prentice Hall (1996)

[31] Lin, J.C.-C., Lin, P.-Y., Chang, T.-M., Lin, T.-K., Weng, Y.-T., Chang, K.-C., Tsai, K.-C., 2012. Development of on-site earthquake early warning system for Taiwan. In: D'Amico S. (Ed.), Earthquake Research and Analysis - New Frontiers in Seismology. InTech. https://doi.org/10.5772/28056.

[32] Lundberg, S.M., Lee, S.I., 2017. A unified approach to interpreting model predictions. In: Guyon, I., Von Luxburg, U., Bengio, S., Wallach, H., Fergus, R., Vishwanathan, S., Garnett, R. (Eds.), Advances in Neural Information Processing Systems 30.

[33] S.K. McBride, A. Bostrom, J. Sutton, R.M. de Groot, A.S. Baltay, B. Terbush, P. Bodin, et al.. Developing post-alert messaging for Shakealert, the earthquake early warning system for the West Coast of the United States of America. Int. J. Disaster Risk Reduct., 50 (2020), Article 101713,

[34] X. Meng, C.A. Goulet. Lessons learned from applying varying coefficient model to controlled simulation datasets. Bull. Earthq. Eng., 21 (2022), pp. 5151-5174,

[35] Molnar, C., 2020. Interpretable Machine Learning. Lulu.com.

[36] J. Münchmeyer, D. Bindi, U. Leser, F. Tilmann. The transformer earthquake alerting model: a new versatile approach to earthquake early warning. Geophys. J. Int., 225 (1) (2021), pp. 646-656,

[37] National Research Institute for Earth Science and Disaster Resilience, 2019. NIED K-NET, KiK-Net. National Research Institute for Earth Science and Disaster Resilience.

[38] A. Páez, D.C. Wheeler. Geographically weighted regression. Int. Encyclopedia Human Geogr., 47 (3) (2009), pp. 407-414,

[39] Roth, A.E., 1988. The Shapley Value: Essays in Honor of Lloyd S. Shapley. Cambridge University Press, Cambridge. https://doi.org/10.1017/CBO9780511528446.

[40] F. Tajima, T. Hayashida. Earthquake early warning: What does “seconds before a strong hit” mean?. Prog. Earth Planet Sci., 5 (1) (2018), p. 63,

[41] A.S. Whittake, M. Kumar, M. Kumar. Seismic isolation of nuclear power plants. Nucl. Eng. Technol., 46 (5) (2014), pp. 569-580,

[42] S. Wu, J.L. Beck, T.H. Heaton. ePAD: Earthquake probability-based automated decision-making framework for earthquake early warning. Comput. Aided Civ. Inf. Eng., 28 (10) (2013), pp. 737-752,

[43] Y. Xiang, N. Farzad, Z. Farzin. Evaluation of natural periods and modal damping ratios for seismic design of building structures. Earthq. Spectra, 36 (2) (2020), pp. 629-646,