Research Proposal#

Surrogate Modeling for Efficient Earthquake Prediction#

Vishal Kamalakrishnan, Kevin Ma, Zachary Russell, Justin Zheng

University of Virginia

Professor Geoffrey Fox

Surrogate Simulation Team 3

5/31/2023 – 7/24/2023

Introduction#

Earthquakes pose a significant threat to human lives and infrastructure, necessitating accurate prediction models for timely responses. Traditional physics-based simulations and monitoring systems, while accurate, often require substantial computational time. This research aims to develop a surrogate model that matches the accuracy of physics-based models while significantly reducing computation time. By utilizing machine learning and data analysis techniques, the study seeks to create a surrogate model for quick seismic activity assessment and earthquake prediction, enabling timely warnings and effective emergency response.

Problem Statement#

The research addresses the need for accurate and computationally efficient earthquake prediction models. By bridging this gap, the study aims to advance industry practices in earthquake forecasting, specifically in risk assessment and location prediction. The research seeks to answer the following research question: How much time can be saved by employing a surrogate model for earthquake prediction compared to typical physics-based or computational models?

Objectives#

Development#

Develop a surrogate model for earthquake prediction achieving accuracy comparable to physics-based models while significantly reducing computation time.

Expected Benefits/Impact:

  • Increased efficiency in earthquake prediction through faster computation.

  • Improved accessibility and usability of earthquake prediction models.

  • Enhanced understanding of earthquake patterns and relationships.

  • Reliable predictions for real-time applications and decision-making.

Validity#

Validate the surrogate model by comparing its predictions with physics-based models using a comprehensive dataset.

Expected Benefits/Impact:

  • Quantitative assessment of the surrogate model’s accuracy.

  • Confidence in its reliability for earthquake prediction.

  • Identification of potential improvements.

Optimization#

Optimize the surrogate model by refining training methodology, input features, and model architecture.

Expected Benefits/Impact:

  • Improved accuracy and reduced biases or errors.

  • Determination of key input features for accurate predictions.

  • Development of an optimized, efficient surrogate model.

Performance#

Assess the generalizability of the surrogate model by evaluating its performance on diverse datasets from different earthquake-prone regions.

Expected Benefits/Impact:

  • Evaluation of model performance across various locations and earthquake characteristics.

  • Validation of its applicability as a versatile earthquake prediction tool.

Methodology#

Data Collection:#

  • Gather relevant earthquake-related data including positions, growth/size, and information about cities and urban districts.

  • Obtain general information such as dates, times, and other relevant factors.

Surrogate Model Training and Testing:#

  • Identify suitable datasets for training the surrogate model.

  • Train the surrogate model using the collected data.

  • Validate the model’s performance by testing it with sample earthquakes.

Configuration of Rivanna BII Clusters:#

  • Determine the necessary configuration for utilizing the Rivanna BII Clusters to process the ETAS Model.

  • Collaborate with the team to access the required processing speeds for optimal performance.

References#

  1. “Earthquake - Magnitude, Seismology, Epicenter | Britannica.” Encyclopedia Britannica, https://www.britannica.com/science/earthquake-geology/Earthquake-magnitude. Accessed 12 July 2023.

  2. Fox, Geoffrey Charles, et al. “Earthquake Nowcasting with Deep Learning.” MDPI, 15 Apr. 2022, www.mdpi.com/2624-795X/3/2/11.

  3. Qian, Jing, et al. “Surrogate-Assisted Seismic Performance Assessment Incorporating Vine Copula Captured Dependence.” Engineering Structures, 2 Mar. 2022, www.sciencedirect.com/science/article/abs/pii/S0141029622002139.

  4. Davidsen, J., et al. Generalized Omori–Utsu Law for Aftershock Sequences in Southern California. 6 Feb. 2015, https://www.dropbox.com/sh/ctovgpynfmrucj0/AACzbvQd4izRMwe3PU_jcfiva?dl=0&preview=Davidsen_et_al_ggv061-5.pdf.

  5. Jalilian, Abdollah, and Jiancang Zhuang. Package “ETAS.” 28 Nov. 2022, https://www.dropbox.com/sh/ctovgpynfmrucj0/AACzbvQd4izRMwe3PU_jcfiva?dl=0&preview=ETAS_Package_Zhuang.pdf.

  6. Hainzl, Sebastian. ETAS-Approach Accounting for Short-Term Incompleteness of Earthquake Catalogs. Feb. 2022, https://www.dropbox.com/sh/ctovgpynfmrucj0/AACzbvQd4izRMwe3PU_jcfiva?dl=0&preview=Hainzl_bssa-2021146.1-2.pdf.

  7. Hainzl, S., et al. Statistical Estimation of the Duration of Aftershock Sequences. 19 Feb. 2016, https://www.dropbox.com/sh/ctovgpynfmrucj0/AACzbvQd4izRMwe3PU_jcfiva?dl=0&preview=Hainzl_et_al_ggw075-3.pdf.

  8. Hardebeck, Jeanne L. Appendix S — Constraining E Pidemic T Ype A Ftershock S Equence (ETAS) Parameters from the U Niform C Alifornia E Arthquake R Upture F Orecast, Version 3 Catalog and Validating the ETAS Model for M Agnitude 6.5 or Greater Earthquakes. https://www.dropbox.com/sh/ctovgpynfmrucj0/AACzbvQd4izRMwe3PU_jcfiva?dl=0&preview=Hardebeck+ETAS.pdf.

  9. Helmstetter, Agnes, et al. Comparison of Short-Term and Time-Independent Earthquake Forecast Models for Southern California. Feb. 2006, https://www.dropbox.com/sh/ctovgpynfmrucj0/AACzbvQd4izRMwe3PU_jcfiva?dl=0&preview=Helm_et_al_90_961_05067.pdf.

  10. Kattamanchi, Sasi, et al. Non - Stationary ETAS to Model Earthquake Occurrences Affected by Episodic Aseismic Transients. 2017, https://www.dropbox.com/sh/ctovgpynfmrucj0/AACzbvQd4izRMwe3PU_jcfiva?dl=0&preview=Kattamanchi_et_al_s40623-017-0741-0.pdf.

  11. Lombardi, Anna Maria. Estimation of the Parameters of ETAS Models by Simulated Annealing. Feb. 2015, https://www.dropbox.com/sh/ctovgpynfmrucj0/AACzbvQd4izRMwe3PU_jcfiva?dl=0&preview=Lombardi_et_al_srep08417.pdf.