3-D Interpolation of Scattered Data

This example of interpolation was done using the Modified Quadratic Shepard's method developed by R.J. Renka (see Multivariate interpolation of large sets of scattered data, R.J. Renka, ACM Trans. on Math. Software 14:2, June 1988, pp. 139-148.) The 3-variable function f(x,y,z) = [1.25+cos(5.4y)] cos(6z) / [6+6(3x-1)2] defined on all 8,000 nodes of a 20x20x20 grid was used to create the true results shown below. Since there are 3 axes for the independent variables (x,y,z), the image illustrated below reflects the actual value of the function f(x,y,z) at each point (x,y,z) as viewed from the y-z plane.  Note: All 2D planes cut from the 3D hypervolume are made at the origin of the 3rd axis.  There are actually 10 possible points along the 3rd axis to observe the relevant 2D plane.  For example, the y-z plane shown below is found at x=0.  There are a total of 10 y-z planes that could be cut along the x-axis.
     

    8000-Node True Results

The following images shown below are examples of the interpolated grid starting with 50, 100, 500, 1000, 2000, and 4000 known nodes randomly extracted from the 8,000 node grid. The value of the function f(x,y,z) at each point is stored in a netCDF file (version 3.3), and a netCDF viewer (Ncview ) is used to display the results as a 2-D surface plot viewed from the y-z plane. With random distribution of nodes, the 2000-node set proved to be the smallest set of known nodes to produce acceptable results.