EGU General Assembly 2020
© Author(s) 2020. This work is distributed under
the Creative Commons Attribution 4.0 License.

Upscaling of geomechanical properties in Discrete Element Method (DEM) models of volcano-tectonics

Claire Harnett1, Eoghan Holohan1, Mark Thomas2, and Martin Schöpfer3
Claire Harnett et al.
  • 1School of Earth Sciences, University College Dublin, Dublin, Ireland
  • 2School of Earth and Environment, University of Leeds, Leeds, UK
  • 3Department for Geodynamics and Sedimentology, University of Vienna, Vienna, Austria

In volcanology, as in other branches of geosciences, uncertainties exist around how well rock properties constrained on the laboratory scale represent those at the field scale. For volcano deformation, scale-related uncertainties are compounded by changes in geomechanical properties as progressive deformation evolves to large strains. Furthermore, such large strain deformation is often localised along large-scale discontinuities. It is therefore difficult to investigate this deformation by using traditional continuum modelling approaches. Here we provide an overview of recent Discrete Element Method (DEM) modelling results as applied to large strain, edifice-scale deformation phenomena, such as lava dome instability and caldera collapse. The DEM is a particle-based numerical modelling approach that enables simulation of strain localisation and highly discontinuous deformation.

Upscaling the geomechanical properties of volcanic rocks from the laboratory to the field can be achieved in DEM models through a calibration process that simulates both the laboratory rock testing and field-scale examples. For lava dome collapse, through comparison of observed and modelled attributes (e.g., displacement, dome growth), we infer that field-scale bulk rock properties (i.e., strength, elastic moduli) are approximately 30% of typical laboratory-scale properties. For caldera collapse, varying the same geomechanical properties produces a range of observed styles of caldera collapse, but the properties required at the edifice scale are approximately a factor of 10 lower than typical laboratory-scale properties. Both the calibration of geomechanical properties and the structural outcomes of DEM simulations, and hence the accuracy of upscaling, are fundamentally dependent on the model resolution, which is a function of both the particle size and distribution. The chosen resolution particularly affects rock strength, fracture toughness, and crack development and propagation. Nonetheless, previously reported discrepancies between seismic and geodetic moments for certain volcano-tectonic events are consistent with the upscaled geomechanical properties in edifice-scale DEM simulations, in that such discrepancies can be explained by a similar-sized reduction in the properties derived from laboratory-scale rock tests.  

How to cite: Harnett, C., Holohan, E., Thomas, M., and Schöpfer, M.: Upscaling of geomechanical properties in Discrete Element Method (DEM) models of volcano-tectonics, EGU General Assembly 2020, Online, 4–8 May 2020, EGU2020-3379,, 2020

Comments on the presentation

AC: Author Comment | CC: Community Comment | Report abuse

Presentation version 2 – uploaded on 08 May 2020
There was an error in the upscaling factor on slide 2, which has now been updated
  • AC1: Response to CC1, Claire Harnett, 08 May 2020

    Hi @Sam,

    Thanks for your question! The particle size in the models is largely chosen on the basis of computation time, and some other constraints within the software (e.g., in the dome models, there must be a minimum of 10 particles across the conduit). The first step is to calibrate the model properties with those from the lab and to do this, we use ~40 mm samples, and a corresponding particle diameter. In order to model field-scale processes, we need to use larger particles to minimise computation time, therefore we run a series of further calibration tests to make sure the desired behaviour is still achieved. I think that using the same particle size in the calibration and the final model helps to minimise the effect of particle size on the overall outcome. In the particular software used here (Particle Flow Code), there have been some studies to show that the elastic constants and uniaxial compressive strength are independent of particle size (Potyondy and Cundall, 2004; Potyondy, 2012), but the tensile strength is dependent on particle size – with this in mind, there is certainly some effect of particle size on the scaling. Although I have not conducted a comprehensive testing of this yet, it is certainly something I plan to look more into in the future.

    Thanks again,


    • CC1: Reply to AC1, Sam Poppe, 08 May 2020

      Thanks for clarifying Claire, looking forward to what comes next!

  • CC2: Questions and answers from the live chat during EGU2020, Michael Heap, 11 May 2020

    Q: When replicating the mechanical behaviour of the laboratory test, I notice that your model does not capture the initial non-linear part of the stress-strain curve, usually associated with pre-existing microcrack closure. Do you see this as a problem when calibrating your model?

    A: Yes, this is because the modelled samples are not pre-populated with cracks so we do not see the initial crack closure seen in the lab. It could be done, but would require a much more complex modelling methodology

    Q: Do you need to apply an upscaling factor to the Young’s modulus as well as the UCS?

    A: Good question! in the Colima models, we apply the same scaling factor to the modulus as to the UCS (this is uniaxial compressive strength). I have not rigorously explored whether this is sufficient or whether a more complex relationship exists between the scaling of modulus and scaling of UCS, but it is certainly key to obtaining realistic field scale models

    Q: Hello, what does UCS mean ? Question 2, what 's the initial (visco elastic ?) model for Colima that calibrates your upscaling factor? Therefore i suppose 1D/2D/3D is also important in upscaling properties?

    A: (Meaning of UCS already answered.) For the upscaling factor in Colima, we simply iterate the mechanical properties of the material until we are able to match modelled and observed dome growth. This is fed into the dome growth model which has a elastic solid outer layer and a fluid core. Indeed, 3D models are what we are trying to work towards, but currently the computational expense does not allow us to do this very efficiently. The next release of the software will allow better parallelisation of the code, such that I hope to be able to run 3D models with greater efficiency

    Q: Interesting models! Are you planning to study the resulting surface deformation generated by these different models?

    A: This is my current work in progress! We are mainly focusing on surface displacement and hoping to also progress to tilt

    Q: Are you planning on sticking with the 2D models?

    A: Indeed, 3D models are what we are trying to work towards, but currently the computational expense does not allow us to do this very efficiently. The next release of the software will allow better parallelisation of the code, such that I hope to be able to run 3D models with greater efficiency

Presentation version 1 – uploaded on 01 May 2020
  • CC1: Comment on EGU2020-3379, Sam Poppe, 07 May 2020

    Hi Claire et al.,

    Sam P. (Penn State) here. Great work you have going right now! My question: how much are the upscaling factors you report (30% for dome growth, 50% for caldera collapse) affected by element diameter in the DEM? So how large is the effect of your choice of unit size (mainly chosen in relation to computation time)?

    Thanks for presenting the work!!!