How can I interpolate missing values (undefined areas) of a 3D surface plot (2D array) using Python?

Scipy has an interpolation module that can do this. Relying on the above (in the question posting), this code can be run in the next cell:

from scipy import interpolate

# integer arrays for indexing
x_indx, y_indx = np.meshgrid(np.arange(0, zs.shape[1]),
                             np.arange(0, zs.shape[0]))

# mask all invalid values
zs_masked = np.ma.masked_invalid(zs)

# retrieve the valid, non-Nan, defined values
valid_xs = x_indx[~zs_masked.mask]
valid_ys = y_indx[~zs_masked.mask]
valid_zs = zs_masked[~zs_masked.mask]

# generate interpolated array of z-values
zs_interp = interpolate.griddata((valid_xs, valid_ys), valid_zs.ravel(),
                                 (x_indx, y_indx), method='cubic')

# finally, plot the data 
fig = plt.figure()
ax = fig.add_subplot(111, projection='3d')

ax.plot_surface(X=xs, Y=ys, Z=zs_interp)

The following graph is returned:

Note that this code has been optimized for readability and comprehensibility rather than memory efficiency. Re-Optimizing this code for memory efficiency is a trivial task that is left up to the reader