Continent-scale seismic hazard mitigation using data-driven analysis of large-scale computer models

Overview

The highest earthquake risk worldwide is posed by complex networks of faults that fracture large regions of the continents. Here, relative movements of tectonic plates result in broad seismic hazard across the 1000-km wide regions of continental crust that are sheared, squeezed or stretched between them.
This hazard is commonly not distributed evenly, but instead manifests in complex spatial patterns of localised faulting and earthquakes (e.g. across the central portion of the Alpine-Himalayan belt, Figure 1). But whilst it is critical for hazard assessment to accurately characterise this complex pattern, it is poorly sampled in space, both by sparse geodetic observations of slow decadal deformation (e.g. from GNSS), and by short historical and instrumental records of earthquakes.
Geodynamic physics-based models offer a way of directly relating sparse geodetic and seismological observations to the underlying ‘true’ continuous deformation field, as an alternative to simply using empirical or kinematic methods of interpolation. Such approaches enable a much deeper understanding of the fundamental driving forces of the system; e.g. to determine not only where the hazard is located, but also why.

This approach has been successfully used in a handful of cases (e.g. Walters et al., 2017), but a single major limitation prevents wide application: numerical geodynamic models of continental deformation are computationally expensive and slow to run, and yet thousands or tens of thousands of model ‘runs’ are needed to explore which combinations of input model parameters (e.g. boundary forces, material properties of the continental lithosphere) can explain the observations of deformation (e.g. Kaus et al., 2016). This limitation essentially limits this approach to overly simplified geodynamic models.
The objective of this project is therefore to design novel computational techniques that allow us to run “non-simplified” simulations for many parameters quickly and apply them to geodynamic models. This will enable investigation of continent-scale seismic hazard in a wide variety of tectonic settings.

Click on an image to expand

Image Captions

Shaded relief map of central portion of the Alpine-Himalayan belt with active faults and seismicity (small coloured ‘beach-balls’), illustrating the heterogeneous pattern of deformation and hazard (Walters et al., 2017).

Methodology

The project will focus on tectonic deformation and seismic hazard in portions of the Alpine-Himalayan mountain belt. Present-day crustal deformation will be assessed through analysis of existing earthquake catalogues and GNSS velocities. The aim is to construct numerical simulations of the tectonic processes that match the observations. To achieve such a match, we typically run many “trials”, i.e. simulations with various material parameters, and from there ‘move into’ the parameter space direction that minimizes the difference between model simulations and available observations. In machine learning, this approach is referred to as stochastic gradient descent.
Physical calculations will be performed using the community code ASPECT, which has been applied to a wide variety of relevant tectonic and geodynamical modeling projects (Heister et al., 2017). ASPECT uses cutting-edge numerical techniques for optimal performance, is extensible to tailor for individual needs, and runs on large supercomputers.
Despite the efficiency and flexibility of ASPECT, simulations are likely to be too computationally expensive to explore parameter spaces by stochastic gradient descent method alone. Therefore, the concept of ‘adaptive mesh refinement’, which is commonly used to refine solutions in physical space dimensions (i.e. x, y. and z; ASPECT uses this approach as well), will be applied to parameter space dimensions too. This novel approach is aimed at significantly reducing computational costs, and therefore allow model calculations to be more complex.

Project Timeline

Year 1

Gaining familiarity with the project through literature review, introduction and training in numerical modelling and programming (if applicable; delivered through Durham’s MISCADA programme), IAPETUS DTP training, and attendance of a first conference.

Year 2

Completion of the basic software tools to be developed, and first major results on the project; preparation for publication of first key results in a peer-reviewed journal.

Year 3

Application of the developed software and methodologies to different tectonic settings and areas; writing of further manuscript for publication.

Year 3.5

Finalizing further publications of research outcomes; thesis completion and submission.

Training
& Skills

The student will become part of a vibrant research culture in the department of Earth Sciences, in which ~70 PhD students work on a wide range of Earth Science research projects, and will closely collaborate with members of the geodynamics research group.
Training will be provided in geodynamical modelling (programming, code development, model setup, and usage) as well as data management of high-performance computing systems. The project is an opportunity for the student to become proficient in computer programming and large dataset analysis, with support from an enthusiastic ASPECT community.
The student will have opportunities to work with other partners in the UK and internationally and they are encouraged to travel to national and international scientific meetings to present results. The student will become a member of the UK’s Centre for the Observation and Modelling of Earthquakes, Volcanoes and Tectonics (COMET), benefitting from the shared expertise of Geosciences staff in several universities.
The student will also become part of the IAPETUS DTP which offers a multidisciplinary package of training focused around meeting the specific needs and requirements of each of our students who benefit from the combined strengths and expertise that is available across our partner organisations.

References & further reading

Heister, T., Dannberg, J., Gassm̦ller, R., Bangerth, W. (2017). High accuracy mantle convection simulation through modern numerical methods РII: realistic models and problems, Geophysical Journal International, vol. 210(2), pp. 833-851.
Kaus, B.J.P., A. A. Popov, T. S. Baumann, A. E. Pusok, A. Bauville, N. Fernandez, M. Collignon (2016). Forward and inverse modelling of lithospheric deformation on geological timescales. Proceedings of NIC Symposium.
Walters, R. J., P. C. England, and G. A. Houseman (2017), Constraints from GPS measurements on the dynamics of the zone of convergence between Arabia and Eurasia, J. Geophys. Res. Solid Earth, 122, 1470-1495.

Further Information

For any information on the project, the tectonics or geodynamics research groups, the department of Earth Sciences or, more generally, matters related to doing a PhD in Durham, please contact Jeroen van Hunen (jeroen.van-hunen@durham.ac.uk).

Apply Now