1

I am lost in all interpolation methods provided by great SciPy and can't find optimal way for my case.

I have millions of XYZ points in matrix which has different intervals between points (cells) and also is rotated. Generally big datasets of points which are somewhere between regular and scattered representing surface which I need to convert to regular grid for further analysis.

I need something fast but creating smooth surface which is respecting values in points. In GIS software I liked the most spline functions but the tools are running too slow and that's where I turned to SciPy. Also linear / Delaunay triangulation is acceptable but I prefer more smoothed surface.

I tried and realy like SciPy Rbf but it dies with larger number of points. Maybe braking file to smaller tiles and than merging back?

Miro
  • 599
  • 10
  • 29

1 Answers1

2

So far the best solution I found is to do it through matplotlib.mlab using griddata with linear interpolation.

import matplotlib.mlab as ml
zi = ml.griddata(x,y,z,xi,yi,interp='linear')

Update: Two additional improvements. I realized that matplotlib.mlab griddata is not really same as scipy.interpolate griddata and the second is better for my case. Also my coordinates are in millions which for big grids is causing big troubles with presicision (for underlying Qhull library) so it is better to shift all coordinates close to origin and after calculation shift back.

from scipy.interpolate import griddata
x -= shift_x
y -= shift_y
zi = griddata((x,y),z,(xi,yi),method='linear')
xi += shift_x
yi += shift_y
Miro
  • 599
  • 10
  • 29