Scribd
Upload
31 views

Interpolation Slides - EFTF

The document discusses implicit modelling techniques for representing geological surfaces and models from observation data. Implicit functions are fitted to data using basis functions or interpolation to define stratigraphic thicknesses anywhere in the model volume. Surfaces can be extracted from the implicit functions by tracing isovalues. The document compares discrete approximation methods that may not exactly fit data, to data-supported interpolation that guarantees an exact fit but can have poor extrapolation. A simple example compares interpolators on noisy surface location data.

Uploaded by

Lachlan Grose
Copyright
© © All Rights Reserved
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
31 views

Interpolation Slides - EFTF

The document discusses implicit modelling techniques for representing geological surfaces and models from observation data. Implicit functions are fitted to data using basis functions or interpolation to define stratigraphic thicknesses anywhere in the model volume. Surfaces can be extracted from the implicit functions by tracing isovalues. The document compares discrete approximation methods that may not exactly fit data, to data-supported interpolation that guarantees an exact fit but can have poor extrapolation. A simple example compares interpolators on noisy surface location data.

Uploaded by

Lachlan Grose
Copyright
© © All Rights Reserved
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
You are on page 1/ 8

Implicit modelling

Approximate implicit function from geological observations (contacts,


strike and dip) using weight combination of basis functions
◦ Data supported
◦ Discrete approach

Implicit function can be evaluated anywhere in the model volume defining


stratigraphic thicknesses
Surfaces can be extracted by tracing isovalues of the implicit function

Calgagno et al., 2008; Chiles


et al., 2004; Cowan et al.,
2003; Frank et, al., 2007;
Hillier et al., 2014; La-junie
et al., 1997; Mallet 1992,14
Surfaces and 3D models
Two ways to represent surfaces
◦ Explicit – surface geometry is directly represented by the support representing the surface. Surface is
only known
◦ Implicit – surface geometry is independent from representation support

Geological model needs the location and geometry of geological objects


◦ Boundary representation (using surfaces)
◦ Directly using implicit functions

If models are a collection of implicit functions, the models do not require discretisation, could be
updated with new datasets.
Implicit modelling constraints

f(X) – f’(X) = 0
Gradient normal constraints

Gradient tangent constraints

Interface
Scalar value
value
constraint
constraint

f(X) – f’(X) = 0
How implicit modelling works

Least squares minimisation Solve linear system of equations


Balance regularisation and observation constraints Exact fit of function and observations
Important technicalities
Objective: fit the function to geological observations – location, gradient
Geological data types are converted into linear equations:
◦ Value/increment constraints constrain the value of the function
◦ Gradient observations constrain the derivative of the implicit function

The value of the implicit function is important – derivative = change of function value in x,y,z
◦ Wider range in function value = steeper gradient

Value of implicit function


Direct or approximation?
Discrete interpolation is an over constrained optimisation problem
◦ Degrees of freedom = resolution of the mesh (Cartesian or tetrahedral)
◦ Misfit to observations + misfit to smoothness constraint
◦ Resulting implicit function is not guaranteed to fit observations exactly
◦ Weighting of data constraints and regularisation can be changed
◦ Can smooth outliers

Data supported interpolation has one degree of freedom per data point  exact interpolation
◦ Combination of fitting a polynomial trend and a local residual
◦ Implicit function guaranteed to fit data
◦ High curvature geometries can require high order polynomials  poor extrapolation away from data
and high curvature local features

Different interpolation algorithms are suited to different problems, most software uses only one.
Simple example
Surface locations sampled from flat plane, value =0

Surface locations sampled from a flat


plane with noise added, value = 10
Comparing interpolators

Discrete: Discrete: Data supported:


Piecewise Linear interpolator Finite difference interpolator Surfe Radial Basis Interpolation

Single isosurface has


topology suggesting rbf is
over fitting

You might also like