Survey
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
www.sci.monash.edu Challenges in Earth Modelling (Plate tectonics, Subduction … ) www.moresi.info L!" Mor#i et al Moresi 1 Plate Tectonics The grand challenge of understanding plate tectonics — “The grand unifying theory of everything in Geology” — the great systematising framework for global data. But it’s not a global theory Continental deformation is not plate like Mantle motion is tangential Early Earth ? And it has insubstantial dynamic underpinnings How does the Earth’s heat engine produce plate tectonics ? How did it begin ? Why do we see no other examples & what should we look for anyway ? Computational modelling and simulation as a key tool What constitutes plate tectonics / subduction ? Integrating observations What does success in modelling look like ? 2 Modelling or Simulation ? What is the difference ? Models are about developing an understanding of basic processes and exploring how these play out in potentially complicated interactions. Simplification Interacting processes Weighting effects / identifying principle components Looking for what can be ignored Simulations are about reproducing what happened and predicting what might happen in given scenario Fidelity to observations Assimilating data constraints “Prediction” or Forecasting Understanding uncertainty 3 Ingredients for global models with plates Combine understanding gained from kinematics and reconstruction with dynamic modeling Kinematics Pattern evolution (“rules of plate tectonics”) Far-field surface velocities Dynamics Mantle evolution “Rigid” plate interiors (near-surface constraint) Appropriate failure models for distinct boundary types (different small-scale physics in each case) Constitutive models for distinct plate boundary types (different small-scale physics in each case) Continental rheology distinct from oceanic Cratonic lithosphere preservation ! 4 Multiple coupled phenomena Rheology / constitutive laws Non-linearities (including plasticity) Elasticity Damage Anisotropy! Magmatism Heat transport Compositional Evolution (crust / continents / deep anti-continents) Rheology Atmosphere / Hydrosphere / Cryosphere Physical / Chemical interchanges Interactions across surfaces / entrainment Geomorphology & Landscapes Formation Contingencies of accretion Heat / Element budgets Core dynamics 5 Integrated Earth modelling Seismic sections and interpretation Evolution of basin structure Basin subsidence & stretching models Surface erosion and sediment transport 3D tomography and seismicity mantle instabilities reconstructed plate margins subduction zone thermal and mechanical evolution simulation & data grid infrastructure 6 Plate tectonics — a multi-scale / multi-physics* problem *In the sense of Earnest Rutherford 7 Ingredients #1 — non-linear softening ! Figure 1: Stress v. Strain-rate relation of equation (7) for ye viscosity of equation (6) with various values of the power-law index. Softening —!Bercovici, 1993 8 Velocity Vrms Viscosity Ingredients # 2 — History / material tracking Width Nusselt Number Lenardic et Kaula, 1994 - Self Lubricated Mantel Convection: models in two dimensiones History dependence —!Lenardic & Kaula, various papers 1993-4 9 Plate boundaries Compare transform boundaries with spreading centres: Macroscopic element deformation not parallel to micro-scale deformation Micro-scale deformation deformation approximately represented by macroscopic element deformation Constitutive law at plate boundary scale is going to be different for each 10 Boundary evolution is not isotropic at macro scale E Kinematic / reconstruction model provide very strong! geometrical “constraints” at the large scale which require that stress and strain-rate are, at the least, not coaxial — i.e. there is a change in constitutive-law symmetry from one scale to the next A D Dynamic model ij,j ij B = g⇢0 ↵T = ⌧ij p ⌧ij = ✓ij (I) ✓ij = 2⌘ C ij ⇧ijkl ✓kl e↵,(I) T,t = T,ii i3 t ij,j e↵,(I) Dij E τ Viscoelastic stress prediction Plastic correction to yield surface ui T,i + Q |D| 11 Strain-dependence of lithospheric deformation This is a simulation of continental crust being ! stretched in response to far field stresses ! imposed by plate motions. At modest strain, the deformation will often localise onto faults which can be very long-lasting structures; very fine scale in width, but with large lateral dimension and relatively weak. time The history dependence of shear deformation is tractable if we use a Lagrangian frame. active "fault" 12 Fixed mesh with moving “particles” Regular Eulerian mesh for ! momentum equation (efficient solvers) Lagrangian reference frame for: Stress Underworld Material point method ! y (!0 ) ! y (") Viscoelastic B A C Material not destined to be in shear band D Shear band material Compositional tracking Stress-history tensor Strain Plastic strain history (scalar / tensor) Finite element formulation robust, versatile very simple to go back and forth between! particle and mesh representation KE = ! KE = ∑ w p BTp (x p )C p (x p )B p (x p ) ΩE B T (x)C(x)B(x)dΩ p 13 Lagrangian History & Efficient Fluid solvers In the material point method we can keep a mesh which is computationally efficient for diffusion-dominated problems (including Stokes flow) and material points — a.k.a. particles — for tracking history variables. ! ! ! ! This is the technique implemented in Underworld (and other similar codes) and leads to a very natural approach to many “difficult” issues in geological thermal / mechanical models (www.underworldproject.org) 14 Lagrangian History & Efficient Fluid solvers In the material point method we can keep a mesh which is computationally efficient for diffusion-dominated problems (including Stokes flow) and material points — a.k.a. particles — for tracking history variables. ! ! ! ! This is the technique implemented in Underworld (and other similar codes) and leads to a very natural approach to many “difficult” issues in geological thermal / mechanical models (www.underworldproject.org) 15 Is subduction always “classical”, “textbook” ? The Australian accretion of the VanDieland microcontinent resulted in the terrane being deeply embedded in the over-riding plate and left largely undisturbed since then. Evidence of rotations in present day structural grain (from potential fields, paleomagnetism and other indicators). Moresi, L., Betts, P. G., Miller, M. S., & Cayley, R. A. (2014). Dynamics of continental accretion. Nature. doi:10.1038/nature13033 16 Congested subduction zones Importance of continental crust Mason et al, 2008, Betts et al, 2012, Moresi et al, 2014 showed how trench motion and! slab configuration are influenced by! buoyant material colliding! with a subduction zone. raised question: how does! subduction continue after! accretion of one terrane ? How does a microcontinent / ! plateau switch to the ! over-riding plate ? This is important at many ! different scales. ! Can we understand! this process better and can ! we apply this to understand places! where accretion is complete ? 17 Subduction — plate scale models with material history What can we learn about subduction from simple models ? Underworld — finite element models with tracking of small-scale physics Highly parallel code for modern petascale machines Open source / based on open libraries (StGermain and PETSc) Checkpointed and so forth Adopted child of CITCOM Moresi, Betts, Miller, Cayley, Dynamics of continental accretion. Nature, 2014, doi: 10.1038/nature13033 18 Model setup using UNDERWORLD Layer 1 density accounts for ~7km oceanic crust (but not phases changes during subduction) Layer 1 yield strength is (very) low to account for (unresolved) near-surface faulting, entrainment of sediments into the plate boundary &c Viscosity is truncated after averaging (to 105 x asthenosphere) Layer 3 has significant strength for 80 Myr old lithosophere. In some models this layer yields too. Continental Ribbon material replaces layer 1 and layer 2. 19 2D thinking is misleading ! 35 Myr In 3D a small buoyant block is easily accreted or eaten by the subduction zone. How about a large one ? 45 Myr 55 Myr 65 Myr 70 Myr 80 Myr How does the slab recover from accretion with break-off / windowing ? What should we look for in the superficial geological record ? What happens for a large terrane / microcontinent ? 20 Building a mountain belt 21 Like a swan, the motion is driven from below Older & Stronger Moresi, Betts, Miller, Cayley, Dynamics of continental accretion. Nature, 2014, doi: 10.1038/ nature13033 Younger & Weaker 22 Deep embedding — and the mechanics of oroclines A’ Over-riding Plate A Oceanic Plate Margin! shortens and bends The slab rolls back in two different directions (doesn’t need to stretch / tear to do this) There is no plate boundary along A-A’ so any convergence not accommodated by rollback results in indentation Pinning at the end of the indenter supplies the lateral loads that result in lateral transport of material along the margin and bending. 23 Where is the arc ? Trajectories during lateral subduction — almost pure rollback creates dangling slab A A’ A 0 A' 19 Myr 100 22 Myr 23 Myr 200 26 Myr 0 Plan View 300 500 1000 400 A Trench parallel location Depth (km) 15 Myr 500 600 2500 3000 3500 A' 2000 2500 3000 2000 2500 3000 3500 4000 4500 4000 4500 Trench perpendicular location 24 Terrane accretion is a witness to bendy subduction 55 Ma 45 Ma RES? PAC YAK TZ SL KULA PAC FAR FAR PAC PAC Hudson Bay Present 20 Ma A AD N CA .S. U PAC JdF PAC JdF PAC PAC 25 Terrane accretion is a witness to bendy subduction Terrane accretion may be complicated but we can still observe: Bro Plate boundary changes ok Range Accretion repeats Docking is messy s Kalta g Fault Ti n tin Yukon-Tanana NFF Denali Oroclines are part of the story Alask a Ra Upland a ng e Fa Fau lt ul t Selwyn basin Fa ul es t F BS g an R TF Ca ss iar der Bor rra ne ika ch Ke CSF KF ugh tro Subduction zones must recover from accretion for repetition to occur. te Z CS een TkF Qu MT arl Ch e ott t ul Fa YF Kootenay terrane F FR Does subduction actually step back (re-initiate from scratch) in a 2D sense or does this re-adjustment require a 3D way of thinking ? SR PF 200 0 km Figure 1. Terranes of the Canadian-Alaskan Cordillera. Inset shows terrane groupings and tectonic realms. Colpron, M., Nelson, J. L., & Murphy, D. C. (2007). Northern 456789$:;$%988<=9>$?@$AB9$C<=<D5<=E#F<>G<=$C?8D5FF98<;$,=>9A$>B?H>$A988<=9$68?7I5=6>$<=D$A9JA?=5J$89<FK>;$*<F9?L?5J$M<>5=<F$>A8<A<$?@$#=J9>A8<F$N?8AB$ #K985J<$ON#MP$9;6;)$"9FHQ=$M<>5=R$H989$=?A$I89S5?7>FQ$5=JF7D9D$?=$A988<=9$K<I>$?@$AB9$C?8D5FF98<$O9;6;)$TB99F98$9A$<F;)$:UU:R;$T9$5=JF7D9$AB9K$B989$ Cordilleran terranes and their interactions through time. GSA Today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attern that repeats in the modern Earth Z.-H. Li et al. / Earth and Planetary Science Letters 380 (2013) 98 111 map of the eastern Alpine Himalayan belt (modi ed from Hatzfeld and Molnar, 2010; Yin, 2010; Xu et al., 2012). Th The red color indicates the continental collision zones. The blue color represents the lateral extrusion blocks in the ov movement of continental plates and the extrusion of the overriding continental rocks in the Eurasia- xed reference fram McClusky et al., 2000; Wang et al., 2001; Vernant et al., 2004; Zhang et al., 2004; Reilinger et al., 2010). MS: Mediterrane 27 The state of the art today Aspect (Manual) Models reproduce localised plate boundaries Can follow evolving geometry Looking at adaptive meshing for small-scales Physics + chemistry Reaching towards coupling ! surface / hydrosphere / atmosphere User friendly ?! ASPECT Advanced Solver for Problems in Earth’s ConvecTi Preview release, version 0.4.pre (generated from subversion: Revision : 2273) Wolfgang Bangerth Timo Heister with contributions by: Markus Bürg, Juliane Dannberg, René Gaßmöller, Thomas Geenen, Eri Kronbichler, Elvira Mulyukova, Cedric Thieulot Underworld 2 Stag Rhea (Burstedde et al) SC 2010 28 The “flying car future” — 20 years Unrecognisable environments — unlimited resources unlimited streams of data unfamiliar ways to interact with models and data Algorithmic priorities will be very different Assume any equations can be solved efficiently at sufficient resolution (?)! ( We are much less neurotic about loop unrolling etc than we were — let’s not look back ! ) Composing at the conceptual level of datasets / modelsets rather than operations “A Model” will be an ensemble of models of today’s definition To propagate errors, uncertainties etc, it is much better to think of a single model being a whole swarm ! of individual realisations We can create metrics and comparison operators with this approach Models (ensembles) will be richly tagged with evolving metadata Google-style searches on models will be trivial! ( We probably couldn’t imagine how to automatically ! search text for arbitrary concepts 20 years ago ) Libraries / Catalogues of models will be possible / searchable given! metrics etc! 29 This slide is blank 30