|
SSP Project Summary
|
 |
Inference of fluid flow at the top of the Earth's liquid core from
geomagnetic data
Student
Mark Madden, University of York
Supervisors
Rob Baxter, EPCC
Kathy Whaler, Department of Geology and Geophysics, University of Edinburgh
Observed changes in the geomagnetic field on the timescale of decades
to centuries are caused by flow in the liquid iron-rich outer
core. Due to its high electrical conductivity, the fluid drags the
magnetic field lines around with it as it moves (the field is 'frozen
in' to the fluid). Thus, we can use the field changes as 'tracers' of
the flow at the core surface. However, there is an element of
ambiguity in the flow so determined, which we reduce by making extra
assumptions about the nature of the flow.
One such assumption is that the flow is steady - this is attractive
computationally, since we then only require a single set of parameters
describing the flow to explain many decades of geomagnetic
data. However, steady flows cannot explain much of the temporal detail
in the data. We can improve the fit to the data considerably without
sacrificing the simplicity of the steady flow by introducing a simple
time dependency - that the core reference frame drifts azimuthally
with respect to our observation frame. Thus, we have two quantities to
solve for - the drift of the core reference frame and the steady flow
within that drifting frame. Previous calculations have used an
interactive approach (assume a drift, find the best-fitting flow in
that drifting frame, update the drift estimate and repeat), but this
may not fully explore parameter space. This project explores Monte
Carlo sampling using a probabilistic procedure as an alternative.
This project builds on work done by Rob Baxter of EPCC under a NERC grant to
Kathy Whaler and in a previous SSP project SS-97-04, and splits into a number
of stages:
- Nesting the parallel code produced by the previous SSP project into a new
code. This code generates the matrix which is to be inverted to find the
flow velocity from the drift velocity. The generation of the matrix is
currently done in parallel by a task farm, the number of tasks
(= number of epochs data is calculated for) is O(100). The matrix
is not particularly large, O(400*400), and there is no need for a parallel
solver to invert the matrix.
- Parallelising over the generation of drift velocities. We envisage
the overall code being a nested task farm. For a particular drift
velocity a group of processors would generate the matrix and then invert it.
The drift velocities themselves will be produced by a Monte Carlo method.
- If time is available the project can be extended in other ways e.g.
by looking at ways of optimising the code or by using data from the period
1840-1990 to obtain a realistic solution.
The final report for this project is available here.
Webpage maintained by
mario@epcc.ed.ac.uk